{"id": "test_0", "sentence1": "In one-shot imitation learning, the agent must perform a task conditioned on one reference example of the task.", "sentence2": "unlike our method, these approach does not easily allow for sequencing skills into longer horizon tasks or composing tasks via arithmetic operations on the latent representation.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_1", "sentence1": "The success of machine learning (ML) during the past decade has relied heavily on the rapid growth of computational power, new techniques training larger and more representative neural networks, and critically, the availability of enormous amounts of annotated data.", "sentence2": "new challenges have arisen with the move from cloud computing to edge computing and Internet of Things (IoT), and demands for customized models and local data privacy are increasing, which raise the question: how can a powerful model be trained for a specific user using only a limited number of local data?", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_2", "sentence1": "classification that aims to distinguish among cat, dog and others.", "sentence2": "in each step t + 1, each class y aggregates prototypes from its neighbors (parents and children) by multi-head attention, and chooses between the aggregated message or the message from itself by a gate g to update its prototype.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_3", "sentence1": "Hence, tasks can be related through the paths on the graph that links their nodes even when they share few output classes.", "sentence2": "we use linear transformation for g(·) and h(·).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_4", "sentence1": "Experiments show that our algorithm achieves a strong performance over several image classification datasets.", "sentence2": "it obtains an error rate of 1.6% for CIFAR-10, 23.9% for ImageNet under mobile settings, and achieves state-of-the-art results on three additional datasets.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_5", "sentence1": "Unlike conventional deep networks that come with a fixed number L of layers, the runtime of DEQ depends strongly on the number of Broyden steps to reach the equilibrium.", "sentence2": "it's challenging to fairly compare the runtimes of implicit-depth models like DEQ with those of corresponding weight-tied deep networks (e.g., using higher depth necessarily takes longer to run).", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_6", "sentence1": "Such weight-tying is generally considered to come with four major benefits: 1) it acts as a form of regularization that stabilizes training and supports generalization; 2) it significantly reduces the model size; 3) it is trivial to show that any deep network can be represented by a weight-tied deep network of equal depth and only a linear increase in width (see Appendix C); and 4) the network can be unrolled to any depth, typically with improved feature abstractions as depth increases [8,18].", "sentence2": "in practice almost all such models (and deep nets in general) are stacked, trained and evaluated by unrolling a pre-determined, fixed number of layers.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_7", "sentence1": "Importantly, we show that DEQ can directly differentiate through the fixed point equations via implicit differentation, which does not require storing any intermediate activation values.", "sentence2": "we can backpropagate through the infinite-depth network while using only constant memory, equivalent to a single layer's activations.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_8", "sentence1": "Unlike a conventional network where the output is the activations from the L th layer, the output of a DEQ is the equilibrium point itself.", "sentence2": "the forward evaluation could be any procedure that solves for this equilibrium point.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_9", "sentence1": "Among the many applications of deep networks, sequence modeling has witnessed continuous advances in model architectures.", "sentence2": "while recurrent networks have long been the dominant model for sequences [21,26,14,34], deep feedforward architectures based on temporal convolutions [49,47,7] and self-attention [48,16,13] have (re-)emerged to claim superior performance on a variety of sequence prediction tasks.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_10", "sentence1": "It is noted that the optimizationbased methods need fine-tuning on the target task, making the classification time consuming.", "sentence2": "our method requires no model updating solves the tasks in an feed-forward manner, which is much faster and simpler than above methods and has better results.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_11", "sentence1": "A straightforward approach is fine-tuning a pretrained model using the few labeled samples from the unseen classes.", "sentence2": "it may cause severe overfitting.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_12", "sentence1": "These approaches generate the parameters of the feature extractor based on the support set and extract the query features adaptively.", "sentence2": "in this way, we can progressively enrich the class features to be more representative and robust.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_13", "sentence1": "There are m machines, each observing n i.i.d sample functions from P .", "sentence2": "in this work, we present an algorithm with O log(mn) bits per message, which we call Multi-Resolution Estimator for Convex landscapes and log mn bits communication budget (MRE-C-log) algorithm.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_14", "sentence1": "The expected loss of this modified algorithm can be shown to still matches the existing lower bounds up to logarithmic factors.", "sentence2": "in the n = 1 regime, the expected error of MRE-C-log algorithm goes to zero as the number of machines increases, while the expected errors of the previously available estimators remain lower bounded by a constant.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_15", "sentence1": "The observations are also in line with the expected error bounds we give in this paper and those previously available.", "sentence2": "in the n = 1 regime, the expected error of MRE-C-log algorithm goes to zero as the number of machines increases, while the expected errors of the previously available estimators remain lower bounded by a constant.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_16", "sentence1": "Conjugate gradients and related MVM-based algorithms [9,13,40] have been used in certain settings throughout the GP literature.", "sentence2": "although exact GPs take longer to train, we find that their speed is comparable to approximate methods at test time.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_17", "sentence1": "There may be some regimes where other approximate methods or combinations of methods outperform these two approximations.", "sentence2": "for SGPR, we perform 100 iterations of Adam with a learning rate of 0.1.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_18", "sentence1": "Many approximate methods have been introduced to improve scalability, relying on mixture-of-experts models [7], inducing points [12,37,39,42], random feature expansions [20,30,47], or stochastic variational optimization [2,16,17,36,46].", "sentence2": "due to the historical intractability of training exact GPs on large datasets, it has been an open question how approximate methods compare to an exact approach when much more data is available.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_19", "sentence1": "Although forming K XX requires O(n 2 ) memory, the result of the MVM K XX v requires only O(n) memory.", "sentence2": "we reduce the memory requirement to O(n) by computing K XX v in separate constant-sized pieces.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_20", "sentence1": "While each iteration of CG requires computing each kernel matrix partition from scratch, the preconditioner is computed once before any iterations of CG are performed.", "sentence2": "it can be efficient to increase the size of the preconditioner to an extent if it reduces the number of CG iterations.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_21", "sentence1": "For n≥100,000, we must use kernel partitioning as discussed above, which significantly increases the training time for exact GPs.", "sentence2": "if the necessary computational resources are available, these experiments show that it may be preferable to train an exact GP to make more accurate predictions in exchange for longer training times.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_22", "sentence1": "A natural question to ask is whether we can consider our approach \"exact\" in light of the fact that CG perform solves only up to a pre-specified error tolerance.", "sentence2": "unlike the approximate methods presented in this paper, the difference between a CG-based model and a theoretical model with \"perfect\" solves can be precisely controlled by this error tolerance.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_23", "sentence1": "Gaussian processes (GPs) are flexible non-parametric models, with a capacity that grows with the available data.", "sentence2": "computational constraints with standard inference procedures have limited exact GPs to problems with fewer than about ten thousand training points, necessitating approximations for larger datasets.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_24", "sentence1": "The proof of Theorem 2.7 closely follows that of [43,Theorem 9].", "sentence2": "the size of the ambiguity set and the regularization parameter are selected using stratified 5-fold cross validation.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_25", "sentence1": "The generic perturbed-iterate framework is given in Section 4.", "sentence2": "aSYNCADA as given in Algorithm 1 maintains the exact global clock t. this option may not be desirable (or available) in certain asynchronous computing scenarios.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_26", "sentence1": "We further use this framework to derive the first asynchronous online and stochastic optimization algorithm with non-box constraints that uses non-Euclidean regularizers.", "sentence2": "we present and analyze HEDGEHOG, the parallel asynchronous variant of the EG algorithm, also known as Hedge in online linear optimization [34,14], 2 In this paper, we do not further consider BCD-based methods, for two main reasons: a) in general, a BCD update may unnecessarily slow down the convergence of the algorithm by focusing only on a single coordinate of the gradient information, especially in the sparse-data problems we consider in this paper (see, e.g., Pedregosa et al.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_27", "sentence1": "The exact way this is done depends on the specific step-size schedule.", "sentence2": "we consider two situations: First, when the step-size is either constant or a simple function of t (or tt in case of ASYNCADA(⇢)), and second, when diagonal ADA-GRAD step-sizes are used.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_28", "sentence1": "Of particular interest are asynchronous lock-free algorithms, where Recht et al. [30] demostrated first that linear speed-ups are possible: They showed that if ⌧ processes run stochastic gradient descent (SGD) and apply their updates to the same shared iterate without locking, then the overall algorithm (called Hogwild!) converges after the same amount of work as serial SGD, up to a multiplicative factor that increases with the number of concurrent processes and decreases with the sparsity of the problem.", "sentence2": "if the problem is sparse enough, this penalty can be considered a constant, and the algorithm achieves linear speed-up.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_29", "sentence1": "Instead, a somewhat weaker goal is to ensure that the price of parallelism is at most a constant factor: that is, the parallel variant needs at most constant-times more updates (or work).", "sentence2": "using t parallel process requires a wall-clock running time that is only O(1/t)-times that of the serial variant.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_30", "sentence1": "When measuring the quality of a large population of generated images, these underrepresented regions have little impact as it is unlikely that too many generated samples land there -even though the hyperspheres may be large, they are sparsely located and cover a small volume of space in total.", "sentence2": "we see that FID favors configurations with high recall (A, F) over the ones with high precision (B, C), and the same is also true for the individual snapshots.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_31", "sentence1": "In order to approximate these quantities directly, we construct adaptive-resolution finite approximations to the real and generated manifolds that are able to answer binary membership queries: \"does sample x lie in the support of distribution P ?\".", "sentence2": "they make effects in the \"null space\" of FID clearly visible.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_32", "sentence1": "As shown in Figure 3, each input node ν ∈ V is represented by concatenating its corresponding semantic embedding a ν with a random Gaussian sample zν .", "sentence2": "early approaches address it as an embedding problem [1,2,6,8,16,24,32,36,39,42,43,48].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_33", "sentence1": "In hyperspherical prototype networks, classification and regression are optimized in the same manner based on a cosine similarity loss.", "sentence2": "both tasks can be optimized not only with the same base network, as is common in multi-task learning [6], but even in the same output space.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_34", "sentence1": "We found that a direct alignment impedes separation, since word embedding representations do not fully incorporate a notion of separation.", "sentence2": "we use a ranking-based loss function, which incorporates similarity order instead of direct similarities.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_35", "sentence1": "As illustrated in Figure 1, a comparison of optimization landscapes without and with preconditioning shows that preconditioning symmetrifies the optimization landscapes and eliminates spurious local minimizers.", "sentence2": "it makes the problem more amendable for optimization.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_36", "sentence1": "As demonstrated in Figure 6, optimizing Huber-loss using vanilla RGD can near perfectly recover both the underlying Bessel PSF and point-sources, producing accurate high resolution image.", "sentence2": "optimizing l4 -loss [21] fails to recover the PSF, resulting in some aliasing effects of the recovered image.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_37", "sentence1": "First, GMF-Q converges after about 10 outer iterations; secondly, as the number of inner iterations increases, the error decreases (Figure 2); and finally, the convergence is robust with respect to both the change of number of states and the initial population distribution (Figure 3).", "sentence2": "the Naive algorithm does not converge even with 10000 inner iterations, and the joint distribution L t keeps fluctuating (Figure 4).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_38", "sentence1": "As |S| and |A| increase, GMF-Q is robust with respect to this increase of dimensionality, while both MF-Q and IL clearly suffer from the increase of the dimensionality with decreased convergence rate and accuracy.", "sentence2": "gMF-Q is more scalable than IL and MF-Q, when the system is complex and the number of players N is large.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_39", "sentence1": "Even Density Functional Theory (DFT) [Hohenberg and Kohn, 1964], a widely used approximation to the equations of quantum mechanics, has trouble scaling to more than a few hundred atoms.", "sentence2": "the majority of practical work in molecular dynamics today falls back on fundamentally classical models, where the atoms are essentially treated as solid balls and the forces between them are given by pre-defined formulae called atomic force fields or empirical potentials, such as the CHARMM family of models [Brooks et al., 1983[Brooks et al., , 2009.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_40", "sentence1": "Every neuron n i corresponds to some subset S i of the atoms.", "sentence2": "each input neuron corresponds to a single atom.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_41", "sentence1": "In principle, quantum mechanics provides a perfect description of the forces governing the behavior of atoms, molecules and crystalline materials such as metals.", "sentence2": "for systems larger than a few dozen atoms, solving the Schrödinger equation explicitly at every timestep is not a feasible proposition on present day computers.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_42", "sentence1": "Our Cormorant neural network has invariance to rotations baked into its architecture in a way that is similar to the physical equations of the previous section: the internal activations are all spherical tensors, which are then combined at the top of the network in such a way as to guarantee that the final output is a scalar (i.e., is invariant).", "sentence2": "to allow the network to learn interactions that are more complicated than classical interatomic forces, we allow each neuron to output not just a single spherical tensor, but a combination of spherical tensors of different orders.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_43", "sentence1": "However it does have a well studied decomposition into irreducibles, called the Clebsch-Gordan decomposition: ) is an 'th order spherical tensor.", "sentence2": "we note that similar Clebsch-Gordan nonlinearities were used in , and that the Clebsch-Gordan product is also an essential part of Tensor Field Networks [Thomas et al., 2018].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_44", "sentence1": "Given T , a bottom-up path is any path connecting a node with one of its ancestors in T .", "sentence2": "we call a backbone path any bottom up path having maximal length.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_45", "sentence1": "A distinctive feature of these algorithms is that they are rather ad hoc in the way they deal with the structure of our problem.", "sentence2": "they cannot be seen as finding the query that splits the version space as evenly as possible, a common approach in many active learning papers (e.g., [12,25,15,16,27,24], and references therein).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_46", "sentence1": "Here, for a given prior P(c * ), the goal of learning is to identify C * either exactly (when λ = 0) or approximately (when λ > 0), while bounding the expected number of queries (x it , x jt ) made to the ground-truth matrix Σ, the expectation being over the noise, and possibly over P(c * ).", "sentence2": "if C is the clustering produced by the algorithm after it stops, we would like to prove upper bounds on E[d H (Σ * , C)], as related to the number of active learning rounds, as well as to the properties of the prior distribution.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_47", "sentence1": "Extending this result to deep ReLU networks requires understanding their behaviour under composition.", "sentence2": "we have ridge functions which vanish on some half space, i.e. colloquially speaking each neuron may “discard half the information” it receives from the previous layer.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_48", "sentence1": "Note, however, that each realization has many different, possibly degenerate, parametrizations.", "sentence2": "a local minimum in the parametrization space needs not correspond to a local minimum in the realization space.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_49", "sentence1": "We proceed by demonstrating how inverse stability opens up new perspectives on the optimization problem which arises in neural network training.", "sentence2": "consider a loss function on the space of continuous functions.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_50", "sentence1": "Moreover, local minima of (8) in the parametrization space must correspond to local minima of ( 9) in the realization space if and only if we have inverse stability.", "sentence2": "furthermore, we show that by optimizing over such restricted sets, it is still possible to learn any function which can be learned by optimization over unrestricted sets.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_51", "sentence1": "Inverse stability is also prevented by redundant directions as the following example illustrates.", "sentence2": "consider a loss function on the space of continuous functions.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_52", "sentence1": "The design of our generative model is inspired by the way humans write articles: Each word should be semantically consistent with not only its surrounding words, but also the entire paragraph/document.", "sentence2": "we assume text generation is a two-step process: A center word is first generated according to the semantics of the paragraph, and then the surrounding words are generated based on the center word's semantics 2 .", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_53", "sentence1": "Since the optimization problem is constrained on the unit sphere, the Euclidean space optimization methods such as SGD cannot be used to optimize our objective, because the Euclidean gradient provides the update direction in a non-curvature space, while the parameters in our model must be updated on a surface with constant positive curvature.", "sentence2": "we need to base our embedding training problem on Riemannian optimization.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_54", "sentence1": "This is probably because negative samples are randomly sampled from the vocabulary and most of them are semantically irrelevant with the center word.", "sentence2": "their ideal embeddings should be orthogonal to the center word's embedding.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_55", "sentence1": "Importantly, previous NAS approaches have focused on searching for E p directly by using |E| as one of the optimization objectives [14].", "sentence2": "1b illustrates one such example.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_56", "sentence1": "Let H be the function class: In contrast, a privacy definition is said to compose sequentially if the privacy properties of algorithms that satisfy it degrade gracefully as the same dataset is used in multiple private releases.", "sentence2": "renyi differential privacy is said to satisfy sequential additive composition -if multiple private algorithms are used on the same dataset, then their privacy parameters add up.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_57", "sentence1": "Let f be a lower semi-continuous convex function such that f(1) = 0, and let P and Q be two distributions over a probability space (Ω, Σ) such that P is absolutely continuous with respect to Q.", "sentence2": "we can show that for any H and for any Γ, (H, Γ)-capacity bounded differential privacy is preserved after post-processing if certain conditions about the function classes hold: Theorem 1.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_58", "sentence1": "Let H 1 , H 2 be two function classes that are convex and translation invariant.", "sentence2": "motivated by these concerns, a recent line of work in machine learning investigates properties that algorithms and methods should possess so that they can automatically guarantee generalization [27,10,6,29].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_59", "sentence1": "We note that there also exists a conceptually simpler smoothing technique based indirect algorithm, which prefixes the tolerance of ε (Appendix D).", "sentence2": "our goal is to find a direct algorithm which does note require prefixing the tolerance at ε.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_60", "sentence1": "Remark 2: Unlike standard AGD for h(y), which only updates y k in the outer-loop, DIAG's outerstep updates both x k and y k thus allowing us to better track the primal-dual gap.", "sentence2": "dIAG's dependence on the condition number L/σ seems sub-optimal and can perhaps be improved if we do not compute Imp-STEP nearly optimally allowing for inexact updates; we leave further investigation into improved dependence on the condition number for future work.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_61", "sentence1": "Now, first order stationary point of a non-smooth nonconvex function is welldefined, i.e., x * is a first order stationary point (FOSP) of a function f (x) if, 0 ∈ ∂f (x * ) (see Definition 3).", "sentence2": "unlike smooth functions, it is nontrivial to define an approximate FOSP.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_62", "sentence1": "Recently, [12] showed linear convergence of gradient descent ascent for strongly-convex-concave problem with bilinear coupling when A has full row rank.", "sentence2": "it has remained an open question if the fast rate of O(1/k 2 ) can be achieved for general strongly-convex-concave minimax problems.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_63", "sentence1": "Additionally, we also examine another quantity as an alternative to distance from initialization: the 2 diameter of the parameter space explored by SGD.", "sentence2": "for a fixed initialization and data distribution, we consider the set of all parameters learned by SGD across all draws of a dataset of size m; we then consider the diameter of the smallest ball enclosing this set.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_64", "sentence1": "This called for a \"rethinking\" of conventional, algorithm-independent techniques to explain generalization.", "sentence2": "it was argued that learning-theoretic approaches must be reformed by identifying and incorporating the implicit bias/regularization of stochastic gradient descent (SGD) [6,35,30].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_65", "sentence1": "For our argument, we will rely on a classifier trained empirically, in contrast to our linear examples where we rely on an analytically derived expression for the learned classifier.", "sentence2": "this section illustrates that the effects we modeled theoretically in the linear classifier are indeed reflected in typical training settings, even though here it is difficult to precisely analyze the learning process.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_66", "sentence1": "Traditional wisdom is that uniform convergence bounds are a bad choice for complex classifiers like k-nearest neighbors because these hypotheses classes have infinite VC-dimension (which motivated the need for stability based generalization bounds in these cases [33,5]).", "sentence2": "consider a scenario where the algorithm generalizes well i.e., for every training set S, h S has zero error on S and has small test error.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_67", "sentence1": "Unfortunately, uniform convergence runs into trouble while dealing with such bad datasets.", "sentence2": "as we can see from the above definition, uniform convergence demands that |L D (h S ) − LS (h S )| be small on all datasets in S δ , which excludes a δ fraction of the datasets.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_68", "sentence1": "Traditional wisdom is that uniform convergence bounds are a bad choice for complex classifiers like k-nearest neighbors because these hypotheses classes have infinite VC-dimension (which motivated the need for stability based generalization bounds in these cases [33,5]).", "sentence2": "this sort of an argument against uniform convergence may still leave one with the faint hope that, by aggressively pruning the hypothesis class (depending on the algorithm and the data distribution), one can achieve meaningful uniform convergence.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_69", "sentence1": "Inspired by our linear model in Section 3.1, we conjecture that the weights w learned on a dataset S can be expressed as w 1 +w 2 , where w T 1 φ(x) dominates the output on most test inputs and induces a simple decision boundary.", "sentence2": "it may be possible to apply uniform convergence on the function w T 1 φ(x) to obtain a small generalization bound.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_70", "sentence1": "Motivated by the seemingly insurmountable hurdles towards developing bounds satisfying all the above five necessary criteria, we take a step back and examine how the underlying technique of uniform convergence may itself be inherently limited in the overparameterized regime.", "sentence2": "we present examples of overparameterized linear classifiers and neural networks trained by GD (or SGD) where uniform convergence can provably fail to explain generalization.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_71", "sentence1": "To make this more concrete, for illustration, consider the high-dimensional linear model that sufficiently wide networks have been shown to converge to [17].", "sentence2": "roughly, these networks can be written as h(x) = w T φ(x) where φ(x) is a rich high-dimensional representation of x computed from many random features (chosen independent of training data).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_72", "sentence1": "It is important to recognize that the second objective (exploration) is important only in so far as it facilitates the first (maximizing reward); there is no intrinsic value in reducing uncertainty.", "sentence2": "exploration should be targeted or application specific; it should not excite the system arbitrarily, but rather in such a way that the information gathered is useful for achieving the goal.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_73", "sentence1": "Concerning worst-case performance, nom attains the lowest cost at the initial epoch, as it does no explicit exploration.", "sentence2": "methods that incorporate exploration achieve better performance at subsequent epochs (and in terms of total cost) due to greater reduction in uncertainty.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_74", "sentence1": "In this section we present: (i) some qualitative insights of our meta-model by showing how it is able to coherently represent a sets of tasks in its latent space, (ii) an illustration of why PROFET helps to obtain statistically meaningful results and (iii) a comparison of various methods from the literature on our new benchmark suite.", "sentence2": "we show results for the following state-of-the-art BO methods as well as two popular evolutionary algorithms: • BO with Gaussian processes (BO-GP) [22].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_75", "sentence1": "Notice that dropping a coordinate j ∈ S from the input samples makes the labels of the resulting training set totally independent from the input.", "sentence2": "dropping j ∈ S results in a training set that is still labeled by a noisy parity.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_76", "sentence1": "Nyström approximation: Unlike the QFF approximation where the basis functions (cosine and sine) do not depend on the data, the basis functions used by the Nyström method are data dependent", "sentence2": "we took the adjusted closing price of 29 stocks from January 4th, 2016 to April 10th, 2019.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_77", "sentence1": "In this paper, we assume smoothness for f induced by the structure of a kernel on X .", "sentence2": "we make the standard assumption of a p.s.d. kernel k : X × X → R such that k(x, x) ≤ 1 for all x ∈ X , and f being an element of the reproducing kernel Hilbert space (RKHS) Hk(X ) of smooth real valued functions on X .", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_78", "sentence1": "The proof is based on constructing a finite subset of \"difficult\" functions in H k (X ).", "sentence2": "we choose f as a uniformly sampled function from a finite set {f1, . . . , fM}, where each fj is obtained by shifting a common function g ∈ Hk(Rd) by a different amount such that each of these has a unique maximum, and then cropping to X = [0, 1].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_79", "sentence1": "We clearly see that for the standard VAE, the variance is low where we have data and arbitrary away from data.", "sentence2": "more precisely, we consider weather data from over 130 years.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_80", "sentence1": "Note that this is a more informative sample, as all observations in the sample are likely to influence the same subset of parameters in θ, effectively increasing the degrees of freedom 4 , hence the quality of variance estimation.", "sentence2": "if the variance network is sufficiently expressive, our Monte Carlo gradients under this sampling scheme are of smaller variation and more sparse.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_81", "sentence1": "On these datasets our model is the best-performing neural network.", "sentence2": "the local log-likelihood is approximately 0 regardless of the observed value y(x 0 ), which should be interpreted as a large entropy of y(x 0 ).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_82", "sentence1": "GPs are robust in settings with a low amount of data, and can model a rich class of functions with few hyperparameters.", "sentence2": "gPs are computationally intractable for large amounts of data and limited by the expressiveness of a chosen kernel.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_83", "sentence1": "Experimentally, we have demonstrated that proposed methods are complementary and provide significant improvements over state-of-the-art.", "sentence2": "on benchmark data we have shown that our method improves upon the test set log-likelihood without improving the RMSE, which demonstrate that the uncertainty is a significant improvement over current methods.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_84", "sentence1": "Closest to our setting is the recent work of Lee et al.  [32], which uses implicit differentiation for quadratic programs in a final SVM layer.", "sentence2": "our formulation allows for adapting the full network for generic objectives (beyond hinge-loss), thereby allowing for wider applications.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_85", "sentence1": "While the broad idea of implicit differentiation is well known, it has not been empirically demonstrated in the past for learning more than a few parameters (e.g., hyperparameters), or highly structured settings such as quadratic programs [4].", "sentence2": "our method meta-trains deep neural networks with thousands of parameters.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_86", "sentence1": "Let Note that the derivative (Jacobian) depends only on the final result of the algorithm, and not the path taken by the algorithm.", "sentence2": "in principle any approach of algorithm can be used to compute Alg i (θ), thereby decoupling meta-gradient computation from choice of inner level optimizer.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_87", "sentence1": "The computational cost of 1 CG step is comparable to 1 inner GD step with the MAML algorithm, since both require 1 hessianvector product (see section C for discussion).", "sentence2": "the computational cost as well as memory of iMAML with 100 inner GD steps is significantly smaller than MAML with 100 GD steps.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_88", "sentence1": "The memory use is also independent of the number of CG iterations, since the intermediate computations need not be stored in memory.", "sentence2": "memory for MAML grows linearly in grad steps, reaching the capacity of a 12 GB GPU in approximately 16 steps.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_89", "sentence1": "This straightforward application of the Laplace mechanism was mentioned at least by Dwork and Smith [2009] and has been widely applied since by several authors [e.g. Zhang et al., 2016, Foulds et al., 2016, Park et al., 2016, Honkela et al., 2018, Bernstein and Sheldon, 2018.", "sentence2": " Foulds et al. [2016]  show that the sufficient statistics perturbation is more efficient than OPS for models where both are applicable.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_90", "sentence1": "The mean absolute difference between the Hellinger squared estimates is It is worth pointing out that although the above statistical properties of the Hellinger distance make it well-suited as a variational objective, it does pose computational issues during optimization.", "sentence2": "to avoid numerically unstable gradient estimation, one must transform Hellingerbased objectives such as Eq. (5).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_91", "sentence1": "The parameters of the intrinsic reward functions are updated to maximize the standard accumulated discounted team return from the environment.", "sentence2": "the objective of the entire procedure is consistent with that of the original MARL problem.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_92", "sentence1": "Our approach builds a connection between reward shaping and critic learning.", "sentence2": "we propose to learn each agent a parameterized individual intrinsic reward function by maximizing a centralized critic.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_93", "sentence1": "In order to figure out what has been learned in the intrinsic reward function, we propose to explicitly visualize these rewards.", "sentence2": "we plot the learned intrinsic reward of each agent at each time step in a complete trajectory during testing.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_94", "sentence1": "We compare LIIR with a number of state-of-the-art MARL methods on battle games in StarCraft II.", "sentence2": "4(b) and (c), at time step 6, all the three agents focus fire on one of the enemy Marine, while agent 1 has the lowest HP; after that, agent 1 still keeps firing instead of running away from the enemies and the intrinsic reward function predicts a low r in 1 , indicating that u 1 = attack is not a good action at that time; finally, agent 1 dies at time step 9 and the corresponding intrinsic reward is very low.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_95", "sentence1": "In this paper, we propose to merge the two directions and learn each agent an intrinsic reward function which diversely stimulates the agents at each time step.", "sentence2": "the intrinsic reward for a specific agent will be involved in computing a distinct proxy critic for the agent to direct the updating of its individual policy.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_96", "sentence1": "Inspired by the concept, we propose to introduce the intrinsic reward design into multi-agent systems to distinguish the contributions of the agents when the environment only returns a team reward.", "sentence2": "we learn each agent a parameterized intrinsic reward function that outputs an intrinsic reward for that agent at each time step to induce diversified behaviors.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_97", "sentence1": "It does not build a feature hierarchy.", "sentence2": "such memories are likely mediated by horizontal or recurrent feedback mechanisms or both.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_98", "sentence1": "HPNet in Frame-Frame scheme performs better than PredNet, suggesting that a feature hierarchy is better than a prediction error hierarchy for long-range prediction.", "sentence2": "hPNet in Frame-Frame scheme does not perform as well as PredRNN++ on long range video prediction.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_99", "sentence1": "Visual neurons' receptive fields are spatiotemporal 3D kernels, rather than 2D.", "sentence2": "we use spatiotemporal block of the input video sequence as input to the neuron in our neural network.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_100", "sentence1": "It has been demonstrated to be effective in explaining the predictive suppression phenomena in the inferotemporal cortex [63].", "sentence2": "predNet only builds a hierarchical representation of errors, where the model of a higher layer learns to predict the prediction errors of the lower layer.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_101", "sentence1": "For HPNet, semantic clustering and decoding accuracy improves progressively as one moves up the hierarchy, from 26% in the first layer (module) to 63% for the top module of the 4-module network.", "sentence2": "the better performance of HPNet in long range video predictions might be attributed to its ability to learn semantically meaningful hierarchical spatiotemporal feature representations and movement to movement relationships (see also [51]).", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_102", "sentence1": "We also performed the same experiment on PredNet and PredRNN++ and found that their corresponding R neurons would also exhibit predictive suppression effect to a certain extent, but much smaller in magnitude.", "sentence2": "we can only claim that the neurophysiological finding is consistent with this genre of self-supervised predictive learning models, though HPnet might be a better approximation.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_103", "sentence1": "These memories, together with the generic statistical priors encoded in receptive fields and connectivity of neurons, serve as internal models of the world for predicting incoming visual experiences.", "sentence2": "it is not clear why early visual cortex needs to be involved in the encoding of spatiotemporal memories and what computational roles it might play.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_104", "sentence1": "What is novel and unanticipated about our finding is that V2 neurons' receptive fields are very local, yet this \"memory effect\" depends on the presence of the global context of the entire movie -reducing the movie aperture from 8 o to 3 o , barely larger than the receptive fields of the individual neurons, would annihilate the predictive suppression effects.", "sentence2": "this predictive suppression effects were not due to adaptation of local receptive field features, but reflected a sensitivity to the memory of the global context of the movies or movement.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_105", "sentence1": "We ask for algorithms that enforce well-studied statistical fairness constraints across two protected populations (we focus on the \"equal opportunity\" constraint of [18], which enforces equalized false positive rates or false negative rates, but our techniques also apply to other statistical fairness constraints like \"statistical parity\" [12]).", "sentence2": "we ask for algorithms that satisfy these constraints (with respect to the unknown underlying distribution) at every round of the learning procedure.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_106", "sentence1": "train a model on one half and predict on the other and vice versa).", "sentence2": "hence, double robustness is used mostly as a variance reduction technique.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_107", "sentence1": "Based on this result, we manage to achieve variance-based regret bounds without the need for variance or moment penalization [15,20,9] used in prior work and which can render a computationally tractable policy learning problem, non-convex.", "sentence2": "our method provides a computationally efficient alternative to the variance penalization when the original ERM problem is convex.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_108", "sentence1": "We empirically evaluate our framework on the personalized pricing application with synthetic data.", "sentence2": "we use simulations to assess our estimator's ability to evaluate and optimize personalized pricing functions.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_109", "sentence1": "We note that we present our theoretical results for the simpler case where the nuisance estimates are trained on a separate split of the data.", "sentence2": "our results qualitatively extend to the case where we use the cross-fitting idea of [5] (i.e. train a model on one half and predict on the other and vice versa).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_110", "sentence1": "However, we show that at least asymptotically and without further assumptions on the functions θ0(z) and Σ0(z), this cannot be the case.", "sentence2": "we show that our estimator achieves what is known as the semi-parametric efficient variance limit for our setting.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_111", "sentence1": "This sequence of operations is illustrated in Figure 2c.", "sentence2": "we used a fixed vocabulary size of 32k, sequence length 1024 and batch size 32.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_112", "sentence1": "GPipe also introduces low communication overhead, given that we only need to pass activation tensors at the partition boundaries between accelerators.", "sentence2": "we can achieve efficient scaling performance even on accelerators without high-speed interconnects.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_113", "sentence1": "Figure 2c assumes partitions are evenly balanced.", "sentence2": "memory requirements and computation flops at different layers are often quite imbalanced.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_114", "sentence1": "We apply synchronous mini-batch gradient descent for training, where gradients are accumulated across all micro-batches in a mini-batch and applied at the end of a mini-batch.", "sentence2": "gradient updates using GPipe are consistent regardless of the number of partitions, allowing researchers to easily train increasingly large models by deploying more accelerators.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_115", "sentence1": "On the one hand, controlling Term I, is more or less standard and closely follows previous such bounds on UCB-type algorithms (e.g., Abbasi-Yadkori et al. (2011));  see Appendix B.2 for details.", "sentence2": "controlling Term II, which we call the regret of safety is more delicate.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_116", "sentence1": "As such, it allows us to quantify tradeoffs between learning the safe set and minimizing the regret.", "sentence2": "we propose Safe-LUCB which is comprised of two phases: (i) a pure-exploration phase that speeds up learning the safe set; (ii) a safe exploration-exploitation phase that optimizes  minimizing the regret.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_117", "sentence1": "For example, when managing demand to minimize costs in a power system, it is required that the operational constraints of the power grid are not violated in response to our actions (these can be formulated as linear constraints that depend on the demand).", "sentence2": "for such systems, it becomes important to develop new bandit algorithms that account for critical safety requirements.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_118", "sentence1": "Here, we present an alternative.", "sentence2": "we propose a variation of Safe-LUCB refered to as generalized safe linear upper confidence bound (GSLUCB).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_119", "sentence1": "However, here we introduce necessary modifications to account for the safety constraint (1).", "sentence2": "we choose the actions with the following two principles.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_120", "sentence1": "Note that for notational simplicity our linear regression model is multiplied by √ n compared to standard scaling in high-dimensional linear regression [BRT09].", "sentence2": "a second contribution is to propose a new knockoffs mechanism, called Conditionally Independent Knockoffs (CIK), which possesses both simple analytic expressions and excellent experimental performance.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_121", "sentence1": "We introduce a simple functional called effective signal deficiency (ESD) of the covariance matrix of the predictors that predicts consistency of various variable selection methods.", "sentence2": "eSD reveals that the structure of the precision matrix plays a central role in consistency and therefore, so does the conditional independence structure of the predictors.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_122", "sentence1": "While some recent work has begun to consider stability properties of neural networks [5,17,19], it has typically done so by (\"softly\") enforcing stability as an additional loss term on the training data.", "sentence2": "they can say little about the stability of the system in unseen states.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_123", "sentence1": "Notably, our proposed approach here is able to learn provably stable systems without solving this (generally hard) problem.", "sentence2": "while it is difficult to find a Lyapunov function that certifies the stability of some known system, we exploit the fact that it is relatively much easier to enforce some function to behave in a stable manner according to a Lyapunov function.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_124", "sentence1": "We also show how such dynamics models can be integrated into larger network systems to learn dynamics over complex output spaces.", "sentence2": "despite the predictive power of these systems, it has been difficult to make formal claims about the basic properties of the learned systems.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_125", "sentence1": "The intuition of the approach we propose in this paper is straightforward: instead of learning a dynamics function and attempting to separately verify its stability via a Lyapunov function, we propose to jointly learn a dynamics model and Lyapunov function, where the dynamics is inherently constrained to be stable (everywhere in the state space) according to the Lyapunov function.", "sentence2": "a damped, rigid n-link pendulum’s state x can be described by the angular position θi and angular velocity θi of each link i.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_126", "sentence1": "For comparison, we also include an example trajectory and resulting frames when the dynamics are modelled without the stability constraint  (i.e. letting f in the above loss be a generic neural network).", "sentence2": "lyapunov functions in deep learning Finally, there has been a small set of recent work exploring the intersection of deep learning and Lyapunov analysis [5,17,19].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_127", "sentence1": "We also show how such dynamics models can be integrated into larger network systems to learn dynamics over complex output spaces.", "sentence2": "we show how to combine the model with a variational auto-encoder (VAE) [11] to learn dynamic \"video textures\" [18].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_128", "sentence1": "Owing to their representational power, neural networks have long been a natural choice for modeling the function f [7,14,13,6].", "sentence2": "when using a generic neural network to model dynamics in this setting, very little can be guaranteed about the behavior of the learned system.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_129", "sentence1": "Although the treatment above seems to make the problem of learning stable systems quite straightforward, the sublety of the approach lies in the choice of the function V .", "sentence2": "as mentioned previously, V needs to be positive definite, but additionally V needs to have no local optima except 0.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_130", "sentence1": "However, because the input polyhedron P 1-ReLU is a hyperrectangle (when projected to X ), it does not capture relational constraints between different x i 's in X (meaning it typically has to substantially over-approximate the set S).", "sentence2": "as expected, the computed result S 1-ReLU of the 1-ReLU method will incur significant imprecision when compared with the S best result.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_131", "sentence1": "We can observe that the volume of the output from 3-ReLU and 2-ReLU is respectively 9 and 7 orders of magnitude smaller than from 1-ReLU.", "sentence2": "as expected, the computed result S 1-ReLU of the 1-ReLU method will incur significant imprecision when compared with the S best result.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_132", "sentence1": "We now show, on a simple example, that the k-ReLU concept can be used to improve the results of state-of-the-art verifiers.", "sentence2": "we illustrate how the output of our verifier kPoly instantiated with 1-ReLU is refined by instantiating it with 2-ReLU.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_133", "sentence1": "Note that P 1-ReLU only contains interval constraints whereas P k-ReLU contains both, the same interval constraints and extra relational constraints.", "sentence2": "any convex relaxation obtained using k-ReLU is typically strictly more precise than a 1-ReLU one.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_134", "sentence1": "Neural networks are being increasingly used in many safety critical domains including autonomous driving, medical devices, and face recognition.", "sentence2": "it is important to ensure they are provably robust and cannot be fooled by adversarial examples [1]: small changes to a given image that can fool the network into making a wrong classification.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_135", "sentence1": "In principle, the high-level policy, πh(g|s), can be trained with any reinforcement learning algorithm, given a suitable way to generate sentences for the goals.", "sentence2": "generating coherent sequences of discrete tokens is difficult, particularly when combined with existing reinforcement learning algorithms.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_136", "sentence1": "In all cases, the agent receives a binary reward only if all constraints are satisfied.", "sentence2": "this makes obtaining meaningful signal in these tasks extremely challenging as only a very small number of action sequences will yield non-zero signal.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_137", "sentence1": "The second approach instead augments the low-level learning with auxiliary rewards that can bring better inductive bias.", "sentence2": "we open the exposition with formalizing the problem of solving temporally extended task with language, including our assumptions regarding the availability of supervision.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_138", "sentence1": "We are interested in learning temporally-extended tasks by leveraging the compositionality of language.", "sentence2": "in addition to the standard reinforcement learning assumptions laid out in Section 3, we also need some form of grounded language supervision in the environemnt E during training.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_139", "sentence1": "Since our low-level policy training is related to goal-conditioned HRL, we can benefit from algorithmic advances in multi-goal reinforcement learning [29,55,4,50].", "sentence2": "we extend the recently popularized goal relabeling strategy [29,4] to instructions, allowing us to relabel based on achieving a language statement that describes a region of state space, rather than relabeling based on reaching an individual state.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_140", "sentence1": "Finally, in the comparison to a bag-of-words representation (BOW), we observe that, while the BOW agent's return increases at a faster rate than that of the language agent at the beginning of traininglikely due to the difficulty of optimizing recurrent neural network in language agent -the language agent achieves significantly better final performance.", "sentence2": "the performance of the BOW agent plateaus at around 8 instructions per episode.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_141", "sentence1": "We use the non-parametric approach of [18] proposed in the context of estimating high-dimensional distributions with sparse covariance structure.", "sentence2": "they use the empirical CDF of the marginal distributions, where m observations are considered.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_142", "sentence1": "We can clearly see that increasing the size of the orientation set in OMP results in a larger angular distance since more dissimilar orientations are included.", "sentence2": "the angular distance of candidate sets chosen by GreedyOrientation decreases fast and then stabilized, which indicates that GreedyOrientation forward selection criterion is defined well so that the best candidate orientations approximate the ground truth are among the immediate ones.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_143", "sentence1": "This optimization, however, is highly under-constrained, with many possible (dense) solutions.", "sentence2": "this objective alone does not enforce a biologically plausible fascicle structure in Φ.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_144", "sentence1": "First of all, we notice that our method in general provides slightly less accurate solutions than the method in [DSSW18], i.e., r e > r e in this case.", "sentence2": " comparing to the brute force algorithm, our method still generates relatively accurate solutions, especially when m is large, e.g., the relative residual percentage w.r.t. the optimal solution is around 1% when m ≈ 16000. ", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_145", "sentence1": "For the Kronecker Product Low-Rank Approximation (LRA) problem, we give an input sparsity O( q i=1 nnz(A i ) + poly(dk/ ))-time algorithm which computes a rank-k matrix B such that down the Kronecker Product A, which is also the running time of state-of-the-art general purpose LRA algorithms [CW13, MM13, NN13].", "sentence2": "our results demonstrate that substantially faster algorithms for approximate LRA are possible for inputs with a Kronecker product structure.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_146", "sentence1": "Our algorithm draws from a large number of disparate sketching techniques, such as the dyadic trick for quickly finding heavy hitters [CM05, KNPW11, LNNT16, NS19], and the precision sampling framework from the streaming literature [AKO11].", "sentence2": "compressing a matrix A to a low-trank approximation yields many of the same benefits as LRA, such as compact representation, fast matrix-vector product, and fast matrix multiplication, and thus is applicable in many of the settings where LRA is used.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_147", "sentence1": "A key contribution of our analysis is to quantify the impact of the part-based structure of the problem on the learning rates of the proposed estimator.", "sentence2": "we prove that under natural assumptions on 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_148", "sentence1": "We identify two main directions for future work: 1) consider settings where the parts are unknown (or \"latent\") and need to be discovered/learned from data; 2) Consider more general locality assumptions.", "sentence2": "for a smaller , such advantage becomes less prominent even when the number of parts is large.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_149", "sentence1": "Efficient inference for gradient approximation: Since R 2 P(M) is a combinatorial space, generally the expensive MCMC algorithm is required for sampling from p (R|T, O) to approximate (16).", "sentence2": "this can be largely accelerated by scrutinizing the logic property in the proposed model.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_150", "sentence1": "In medical trials [11], to identify the best drug for a disease, one can conduct tests in rounds, such that each round involves testing multiple candidate drugs on multiple clinical subjects (e.g., mice) simultaneously.", "sentence2": "after each round of testing, there is typically a waiting time (e.g., days) before the effects of drugs become observable to guide the design of the next round of testing.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_151", "sentence1": "According to Lemma 2, in each iteration, we can bound the total number of arms in S r+1 without double test.", "sentence2": "we go through the algorithm first and then explain why.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_152", "sentence1": "In many real applications, not only the query complexity but also the round complexity need to be optimized.", "sentence2": "another metric was considered in [3], where the identified k arms can have at most ke regret in total.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_153", "sentence1": "Default parameter is set as: δ  = 0.1.", "sentence2": "directly using such indicator without double test, the indicator may be positively biased.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_154", "sentence1": "Then encoding them with the same representation h a = h b can later help the agent extrapolate that if m a is associated with a previously unseen sensory state s a , then m b will also be.", "sentence2": "we propose to train a neural network to perform sensorimotor prediction, and to analyze how it learns to encode motor states depending on the type of exploration that generates the sensorimotor data.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_155", "sentence1": "For example, by taking fewer gradient steps to compute adversarial examples during training.", "sentence2": "this can produce models which are robust against weak attacks, but break down under strong attacks -often due to gradient obfuscation.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_156", "sentence1": "Adversarial training is an effective methodology to train deep neural networks which are robust against adversarial, norm-bounded perturbations.", "sentence2": "the computational cost of adversarial training grows prohibitively as the size of the model and number of input dimensions increase.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_157", "sentence1": "We perform a large scale evaluation of existing methods for adversarially robust training under consistent, strong, white-box attacks.", "sentence2": "the work presented here is a regularization technique which encourages the loss function to be well approximated by its linear Taylor expansion in a sufficiently small neighbourhood.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_158", "sentence1": "In other words, the absolute difference between these two values, is an indicator of how linear the surface is.", "sentence2": "we consider the quantity to be a measure of how linear the surface is within a neighbourhood B( ).", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_159", "sentence1": "The cost of solving Eq (3) is dominated by the cost of solving the inner maximization problem.", "sentence2": "the inner maximization should be performed efficiently to reduce the overall cost of training.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_160", "sentence1": "This non-linearity prevents gradient based optimization methods from finding an adversarial perturbation within a small number of iterations [4,24].", "sentence2": "if the loss surface was linear in the vicinity of the training examples, which is to say well-predicted by local gradient information, gradient obfuscation cannot occur.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_161", "sentence1": "The resulting model trained using LLR degrades gracefully in terms of adversarial accuracy when we increase the strength of attack, as shown in Fig 3 .", "sentence2": "fig 3a shows that, for CIFAR-10, when the attack changes from Untargeted to Multi-Targeted, the LLR's accuracy remains similar with only a 2.18% drop in accuracy.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_162", "sentence1": "The corresponding adversarial loss is given by (x + δ).", "sentence2": "we also see similar trends in accuracy in Table 2.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_163", "sentence1": "The original policy gradient theorem is on-policy and used to optimize the on-policy objective.", "sentence2": "in many cases, we would prefer to learn off-policy to improve data efficiency (Lin, 1992) and exploration (Osband et al., 2018).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_164", "sentence1": "To learn c via stochastic approximation, Hallak and Mannor (2017) propose the COP operator.", "sentence2": "the COP operator does not have a unique fixed point.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_165", "sentence1": "By contrast, evaluating the alternative life objective requires samples from the stationary distribution of the target policy, to which the agent does not have access.", "sentence2": "it is Jπ, not Jµ, that correctly reflects the deploy-time performance of π, as the behavior policy will no longer matter when we deploy the off-policy learned π in a continuing task.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_166", "sentence1": "The data generated by µ is significantly different from any meaningful policy in those tasks.", "sentence2": "this setting exhibits a high degree of off-policyness.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_167", "sentence1": "The data generated by µ is significantly different from any meaningful policy in those tasks.", "sentence2": "our experiments aim to answer the following questions.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_168", "sentence1": "Gradient TD methods are true stochastic gradient methods and enjoy convergence guarantees.", "sentence2": "they are usually two-time-scale, involving two sets of parameters and two learning rates, which makes it hard to use in practice (Sutton et al., 2016).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_169", "sentence1": "Under this task specification (White, 2017), the optimal policy under the alternative life objective Jπ, which is equivalent to the average reward objective as γ and i are constant, is to stay in the outer circle", "sentence2": "to maximize Jµ, the agent prefers the inner circle", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_170", "sentence1": "In a recent paper, Li et al. [2019] study a harder variant of bandits, where the delays d t remain unknown.", "sentence2": "if an action is played at time s and then more times in between time steps s and s + d s , the learner cannot tell which specific round the loss observed at time s + d s refers to.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_171", "sentence1": "In addition to photoreceptors and bipolar cells in the retina [17], ribbon synapses are featured in many other sensory systems, such as in auditory hair cells and the vestibular system [18].", "sentence2": "our proposed Bayesian framework links stimulus-response modeling to a biophysically inspired, mechanistic model of the ribbon synapse.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_172", "sentence1": "Many mechanistic models in computational neuroscience only provide means to simulate data and do not yield an explicit likelihood function.", "sentence2": "their parameters cannot be inferred easily.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_173", "sentence1": "In principle, inference via rejection sampling could be applied, but is often inefficient.", "sentence2": "recently proposed methods use parametric models (like a mixture of Gaussian) to approximate the posterior over several sampling rounds [6].", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_174", "sentence1": "The activity of sensory neurons is noisy -a central goal of systems neuroscience has therefore been to devise probabilistic models that allow to model the stimulus-response relationship of such neurons while capturing their variability [1].", "sentence2": "linear-nonlinear (LN) models and their generalizations have been used extensively to describe neural activity in the retina [2,3].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_175", "sentence1": "In contrast to purely statistical models, the parameters of the LNR model are readily interpretable in terms of properties of the ribbon synapase.", "sentence2": "we show that its parameters can be fitted efficiently on synthetic data and two-photon imaging measurements.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_176", "sentence1": "In Section H, we provide a sketch of this argument and obtain a bound for relu networks that is polynomial in hidden layer and Jacobian norms and inverse preactivations.", "sentence2": "it is not obvious how to adapt the argument of Nagarajan and Kolter [2019] to activation functions whose derivatives are not piecewise-constant.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_177", "sentence1": "Now, following the standard technique of bounding Rademacher complexity via covering numbers, we can obtain generalization error bounds for augmented loss.", "sentence2": "neural nets are complex, so there remain many other data-dependent properties which could potentially lead to better generalization.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_178", "sentence1": "More recently, a semi-supervised deep generative model for attributed network embedding has been proposed [14], which applies the generative adversarial nets (GANs) to generates fake samples in low-density areas in networks and leverages clustering property to help classification.", "sentence2": "it still learns embeddings for nodes only.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_179", "sentence1": "Many efforts have been paid to co-embedding multiple entities for heterogeneous systems in a fully unsupervised learning procedure [24,30].", "sentence2": "to our knowledge, no model in the literature can learn embeddings for heterogeneous data with multiple entities in a semi-supervised way.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_180", "sentence1": "Is it beneficial to adapt the auxiliary task weight based on its longer term effect, i.e. N-step update (Section 4.2) compared to the 1-step update (Section 4.1)?", "sentence2": "for all the manipulation environments, the goal is an RGBD image with objects in the desired configuration.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_181", "sentence1": "Transferring knowledge from other auxiliary tasks is a powerful tool for improving the learning efficiency.", "sentence2": "the usage of auxiliary tasks has been limited so far due to the difficulty in selecting and combining different auxiliary tasks.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_182", "sentence1": "Deep reinforcement learning has enjoyed recent success in domains like games [1,2], robotic manipulation, and locomotion tasks [3,4].", "sentence2": "most of these applications are either limited to simulation or require a large number of samples collected from real-world experiences.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_183", "sentence1": "Previous work manually tunes the auxiliary task weights w i [9].", "sentence2": "a number of issues arise when we try to scale the number of auxiliary tasks.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_184", "sentence1": "However, using gradient-based optimization with only the main task gradient ∇θLmain is often slow and unstable, due to the high variance of reinforcement learning.", "sentence2": "auxiliary tasks are commonly used, especially for image based tasks, to help to learn a good feature representation.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_185", "sentence1": "The supervised baselines use a training set of ground-truth images and generate observations with random parameters on the fly in each epoch, to create a much larger number of effective image-measurement pairs.", "sentence2": "our method is trained with only two measurements per image from the same training set (but not the image itself), with the pairs kept fixed through all epochs of training.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_186", "sentence1": "We begin with the simpler case of non-blind estimation, when the parameter θ for a given measurement y is known, both during inference and training.", "sentence2": "we focus on a more general class of observation models, which requires injecting the measurement process in loss computation.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_187", "sentence1": "However, unlike shallow networks, the training process of RNN often runs into the trouble of vanishing or exploding gradient [53].", "sentence2": "the value of the gradient becomes exponentially small or large in the time horizon, even when the training objective is still constant.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_188", "sentence1": "One may also want to study whether RNN can be trained close to zero test error.", "sentence2": "unlike feedforward networks, the training error, or the ability to memorize examples, may actually be desirable for RNN.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_189", "sentence1": "It is expected that if one could sample a training set ad libitum and measure the learner's true performance over all data, such behavior disappears.", "sentence2": "for the log-likelihood, we basically prove the same: there are standard learners for which the (negative) log-likelihood is not weakly (Z , l, N)-monotonic for any N.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_190", "sentence1": "The hinge loss is appropriate only for the classification setting.", "sentence2": "it should be clear that this paper does not get to the bottom of the learning-curve issue.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_191", "sentence1": "Around the same time, Opper and Kinzel [1996] showed that in the context of neural networks a similar behavior is observed for small samples.", "sentence2": "the error rate for the single layer perceptron is demonstrated to increase when the training set size goes towards the dimensionality of the data [Opper, 2001].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_192", "sentence1": "First of all, we show that nonmonotonic behavior can occur in the setting where the complexity of the learner is small compared to the training set size.", "sentence2": "the reported behavior is not due to jamming or peaking.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_193", "sentence1": "The hinge loss is appropriate only for the classification setting.", "sentence2": "though the rest of the setup remains the same, outputs are limited to the set Y = {−1,+1}  for the hinge loss.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_194", "sentence1": "In other words, there is no discrepancy between the objective used during training and the loss used at test time.", "sentence2": "possibly odd behavior cannot be explained by dipping.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_195", "sentence1": "It is expected that if one could sample a training set ad libitum and measure the learner's true performance over all data, such behavior disappears.", "sentence2": "if one could indeed get to the performance in expectation over all test data and over all training samples of a particular size, performance supposedly improves with more data.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_196", "sentence1": "What this work does convey is that learning curves can (provably) show some rather counterintuitive and surprising behavior.", "sentence2": "we have demonstrated that least squares regression, regression with the absolute loss, linear models trained with the hinge loss, and likelihood estimation of the variance of a normal distribution can all suffer from nonmonotonic behavior, even when evaluated with the loss they optimize for.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_197", "sentence1": "Most optimization methods for solving problems with the Schatten-p norm perform SVD on X at every iteration, with time complexity of O(m 2 n) (supposing m ≤ n) [21,22].", "sentence2": "the natural algorithm to minimize FGSR 2/3 and FGSR 1/2 does not use the SVD, as the regularizers are simple (not spectral) functions of the factors.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_198", "sentence1": "Previous work addresses this issue by adding a variance regularizer [17,22,18] and clipping/truncating the importance weight [6,22].", "sentence2": "the variance regularizer is challenging to use in interactive learning when data arrives sequentially, and it is unclear how the clipping/truncating threshold should be chosen to yield good theoretical guarantees.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_199", "sentence1": "In this set of experiments, we inject faults to CNN models and report the accuracy drops of CNN models protected using different strategies.", "sentence2": "let X be a floating-point tensor and Xq be the 8-bit quantized version.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_200", "sentence1": "Because of its flexibility, RL can encode such a vast array of different problem settings -many of which are entirely intractable.", "sentence2": "it is crucial to understand what conditions enable an RL agent to effectively learn about its environment, and to account for the success of RL methods in practice.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_201", "sentence1": "In inference, to realize the distribution in Equation (5), Ct ∪ Dt is provided like in GQN (see Appendix C.2).", "sentence2": "we see this through the poorer reconstruction and generation quality in the 3D tasks.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_202", "sentence1": "By taking the meta-learning framework, Neural Processes learn to learn a stochastic process quickly from observations while experiencing multiple tasks of stochastic process modeling.", "sentence2": "in Neural Processes, unlike typical neural networks, learning a function is fast and uncertaintyaware while, unlike Gaussian processes, prediction at test time is still efficient.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_203", "sentence1": "Although SSMs have good properties like modularity and interpretability due to the Markovian assumption, the closed-form solution is only available for simple cases like the linear Gaussian SSMs.", "sentence2": "in many applications, SSMs show difficulties in capturing nonlinear non-Markovian long-term dependencies  (Auger-Méthé et al., 2016).", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_204", "sentence1": "In general, the use of PD improves generation quality in all the explored cases.", "sentence2": "the action at each time-step is chosen uniformly.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_205", "sentence1": "In task (a), we are interested in how the transition model generalizes over the time steps.", "sentence2": "we provide context points only in the first 10 time-steps out of 20.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_206", "sentence1": "In GQN, query x corresponds to a camera viewpoint in a 3D space, and output y is an image taken from the camera viewpoint.", "sentence2": "the problem in GQN is cast as: given a context set of viewpoint-image pairs, (i) to infer the representation of the full 3D space and then (ii) to generate an observation image corresponding to a given query viewpoint.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_207", "sentence1": "Neural Processes combine the strengths of neural networks and Gaussian processes to achieve both flexible learning and fast prediction in stochastic processes.", "sentence2": "a large class of problems comprises underlying temporal dependency structures in a sequence of stochastic processes that Neural Processes (NP) do not explicitly consider.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_208", "sentence1": "The term \"pruning\" implies the dropping of connections by setting weights to zero, and these weights are thought of as unimportant.", "sentence2": "if the value of zero for the pruned weights is not important to the performance of the network, we should expect that we can set pruned weights to some other value, such as leaving them frozen at  their initial values, without hurting the trainability of the network.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_209", "sentence1": "This yields fixed-cost adaptation mechanisms, and enables greater sharing across training tasks.", "sentence2": "it may under-fit if the function approximation is not sufficiently flexible.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_210", "sentence1": "Whereas previous work on CNPs has focused on homogeneous regression and classification datasets and fairly simple models, here we study multiple heterogeneous classification datasets and use a more complex model to handle this scenario.", "sentence2": "whereas the original CNP approach to classification required pre-specifying the number of classes in advance, CNAPS handles varying way classification tasks, which is required for e.g.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_211", "sentence1": "However, the dimensionality of this subspace is very low compared to the dimensionality of PreResNet-164 CIFAR-10→STL-10  the weight space, and we can not guarantee that SWAG variance estimates are adequate along all directions in weight space.", "sentence2": "we would expect SWAG to under-estimate the variances along random directions, as the SGD trajectory is in a low-dimensional subspace of the weight space, and a random vector has a close-to-zero projection on this subspace with high probability.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_212", "sentence1": "Bayesian methods provide a natural probabilistic representation of uncertainty in deep learning [e.g., 3, 24, 5], and previously had been a gold standard for inference with neural networks [38].", "sentence2": "existing approaches are often highly sensitive to hyperparameter choices, and hard to scale to modern datasets and architectures, which limits their general applicability in modern deep learning.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_213", "sentence1": "We note that the combination of SWA and temperature scaling presents a competitive baseline.", "sentence2": "unlike SWAG it requires using a validation set to tune the temperature; further, temperature scaling is not effective when the test data distribution differs from train, as we observe in experiments on transfer learning from CIFAR-10 to STL-10.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_214", "sentence1": "For the multi-view representation, CPM-Nets jointly considers multi-view complementarity and class distribution, making them mutually improve each other to obtain the representation reflecting the underlying patterns.", "sentence2": "the encoded latent representation from observations is complete and versatile thus promotes the prediction performance, while the clustering-like classification schema in turn enhances the separability for latent representation.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_215", "sentence1": "For each observation, we retrain the classifier using all past observations-a step requiring samples from the two populations.", "sentence2": "we predict with a uniform distribution (or P (L) when known) until at least one instance of each label has been observed.)", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_216", "sentence1": "In this work, we examined the inductive bias of CNNs, and challenged the accepted view that FCNs are unable to generalize as well as CNNs on visual tasks.", "sentence2": "we showed that the CNN prior is mainly useful during the early stages of training, to prevent the unconstrained FCN from falling prey of spurious solutions with poor generalization too early.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_217", "sentence1": "The practical interest of our protocol is limited by the very large size of the highly sparse eFCN.", "sentence2": "it offers interesting insights into the persistence of architectural bias under stochastic gradient dynamics.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_218", "sentence1": "We then use these approximate gradient oracles to devise approximate gradient descent algorithms.", "sentence2": "we use the stable manifold theorem to prove that zero order methods can almost surely avoid saddle points.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_219", "sentence1": "While local optima may be satisfactory for some applications in machine learning [CHM + 15], saddle points can make high dimensional non convex optimization tasks significantly more difficult [DPG + 14,SQW18].", "sentence2": "researchers have focused their efforts on functions possessing the strict saddle property.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_220", "sentence1": "A recent line of work has shown that gradient descent and variations of it can actually converge to SOSPs .", "sentence2": "[LPP + 19] shows that gradient descent starting from a random point can eventually converge to SOSPs with probability one.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_221", "sentence1": "In the previous subsections we provided sufficient conditions for approximate gradient descent to avoid strict saddle points.", "sentence2": "intuitively, SMT formalizes why convergence to unstable fixed points is unlikely starting from a local region around an unstable fixed point.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_222", "sentence1": "The key insight provided by the SMT is that the all the initialization points that eventually converge to an unstable fixed point lie in a low dimensional manifold.", "sentence2": "to obtain a stronger result we have to understand how SMT restricts the dimensionality of this stable manifold for a fixed h0 .", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_223", "sentence1": "In this case our approximate gradient cannot guarantee a substantial decrease of f .", "sentence2": "we know that the Hessian has a substantially negative eigenvalue and therefore a direction of steep decrease of f must exist.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_224", "sentence1": "In all our experiments, we assume that the weightfunction comes from the class F of mixture of m 2D non-negatively weighted Gaussians as described in the end of Section 3.2.", "sentence2": "the persistent homology is one of the most important developments in the field of topological data analysis, and there have been fundamental developments both on the theoretical front (e.g, [23,10,13,8,14,5]), and on algorithms / implementations (e.g, [43,4,15,20,29,3]).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_225", "sentence1": "There are natural distances defined for persistence diagrams, including the bottleneck distance and the Wasserstein distance, both of which have been well studied (e.g, stability under them [16,18,14]) with efficient implementations available [27,28].", "sentence2": "to facilitate downstream machine learning tasks, it is desirable to further map the persistence diagrams to another \"vector\" representation.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_226", "sentence1": "For example, for the persistence diagrams computed from atomic configurations of molecules, features with small persistence could capture the local packing patterns which are of utmost importance and thus should be given a larger weight; while in many other scenarios, small persistence leads to noise with low importance.", "sentence2": "the 10 * 10-fold nested cross validation are applied to evaluate our algorithm: There are 10 folds in outer loop for evaluation of the model with selected hyperparameters and 10 folds in inner loop for hyperparameter tuning.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_227", "sentence1": "They still lack the investigations on synthesizing the images in a more controllable way, like finely manipulating the visual appearance of every object.", "sentence2": "to generate the images with preferred objects and rich interactions, we propose a semi-parametric method, PasteGAN, for generating the image from the scene graph and the image crops, where spatial arrangements of the objects and their pair-wise relationships are defined by the scene graph and the object appearances are determined by the given object crops.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_228", "sentence1": "In order to integrate the objects in the expected way defined by the scene graph as well as maintaining the visual appearance of the objects, we designed a Crop Refining Network as well as an attentionbased Object-Image Fuser, which can encode the spatial arrangements and visual appearance of the objects as well as their pair-wise interactions into one scene canvas.", "sentence2": "it can encode the complicated interactions between the objects.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_229", "sentence1": "With such algorithms, everyone can become an artist: you just need to define the objects and how they interact with each other, and then the machine will produce the image following your descriptions.", "sentence2": "it is a challenging problem as it requires the model to have a deep visual understanding of the objects as well as how they interact with each other.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_230", "sentence1": "A good crop is not simply matching the category, but also of the similar scene.", "sentence2": "to retrieve the good crop, we should also consider the context information of the object, i.e. e. the entire scene graph it belongs to.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_231", "sentence1": "Policies that aim for high empowerment can lead to complex behavior, e.g. balancing a pole in the absence of any explicit reward signal [23].", "sentence2": "the implicit dependency of p on the optimization argument π empower renders the problem non-trivial.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_232", "sentence1": "Note that our SAC baseline is comparable with the SAC from [22] on Hopper-v2, Walker2d-v2, Ant-v2 and Humanoid-v2 after 5 • 10 5 steps (the SAC from [21] uses the earlier v1-versions of Mujoco and is hence not an optimal reference).", "sentence2": "there is a discrepancy on HalfCheetah-v2.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_233", "sentence1": "This motivates regarding Eq. (4) as a regularization term for learning H, which guides the learning representation process, forming the cluster structures.", "sentence2": "visualization analysis illustrates the effectiveness of cluster-specific temporal representations and demonstrates the robustness of the learning process, even if K-means makes mistakes.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_234", "sentence1": "To further analyze the performance, we perform a pairwise comparison for each method against DTCR.", "sentence2": "we conduct the Wilcoxon signed rank test [36] to measure the significance of the difference.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_235", "sentence1": "DTCR integrates temporal reconstruction and the K-means objective into a seq2seq model.", "sentence2": "dTCR adapts bidirectional Dilated recurrent neural networks [17] as the encoder, enabling the learned representation to capture the temporal dynamics and multi-scale characteristics of time series.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_236", "sentence1": "Note that USSL depends on pseudo-labels to guide the learning, while there is no mechanism to reduce the negative impact when mistakes occur in the pseudo-labels.", "sentence2": "dTCR is capable of correcting mistakes with the help of temporal reconstruction (for analysis see Section 4.3.3).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_237", "sentence1": "Zakaria et al. [24] proposed u-shapelet to learn local patterns.", "sentence2": "the number of selected time steps is bα × Tc, where α ∈ (0, 1] is a hyper-parameter we set to 0.2.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_238", "sentence1": "More importantly, we establish the statistical-computational barrier under an oracle computational model [16,18,19,54], which is an abstraction of computations made by algorithms that interact with data.", "sentence2": "we study the signal detection problem where the link function is defined as a continuous interpolation of two link functions of different types.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_239", "sentence1": "Scale estimation in object tracking is challenging and existing methods mainly enumerate possible scales [2,46] and select the optimal one.", "sentence2": "the scale can be estimated by our proposed model.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_240", "sentence1": "Our model performs favorably against the self-supervised state-of-the-art methods.", "sentence2": "our model outperforms Wang et al. [52] by 13.3% in J and 16.6% in F. and is even 6.9% better in J and 4.1% better in F than the ResNet-18 model [13] trained on ImageNet [9] with classification labels.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_241", "sentence1": "For dense correspondence, the inter-frame affinity needs to be sparse to ensure one-to-one mapping.", "sentence2": "it is challenging to model a sparse matrix in a deep neural network.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_242", "sentence1": "Results are shown in Figure 2 (a).", "sentence2": "the following theorem states that the sparsity constrained optimization problem in (2) is hard even to approximate, in the sense that no deterministic approximation algorithm exists that solves it in polynomial time.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_243", "sentence1": "We have shown in the proof of Theorem 2 that there always exist extreme examples that are hard to solve.", "sentence2": "in the most general sense, and without further assumptions, one can find pathological cases which make the problem hard.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_244", "sentence1": "The projection (2) is iteratively solved in IHT (step 5 in Algorithm 1).", "sentence2": "for the algorithm to be practical, it is important to study the tractability of the projection step.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_245", "sentence1": "Our result can be a significant improvement as m grows.", "sentence2": "in our setting the individual machines need not be able to obtain any estimate of v i , but the corresponding average is still accurate.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_246", "sentence1": "Linear networks are simple since they can only represent linear transformation, but they preserve one of the most important aspects of deep neural networks, the layered structure.", "sentence2": "analysis of linear networks will be helpful for understanding nonlinear cases.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_247", "sentence1": "We observe that the blue curve takes a long time in the neighborhood of saddle point (0, 0), however the red curve does not.", "sentence2": "zAS is equivalent to initializing all the residual blocks and the output layer with zero in a linear residual network.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_248", "sentence1": "[10] provides a simpler proof of this result for deep residual networks, and shows that the he Polyak-Łojasiewic condition is satisfied in a neighborhood of a global minimum.", "sentence2": "these results do not imply that gradient descent can find global minima, and also cannot tell us the number of iterations required for convergence.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_249", "sentence1": "For the case of known order of dying, we propose Algorithm 2, Hedge-Perm-Known (HPK), which is slightly different than HPU.", "sentence2": "the weight redistribution (when an expert dies) and initialization of coefficient h i,1 is different.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_250", "sentence1": "For example, in online advertising, an ad broker has contracts with their providers and these contracts may expire in an order known to Learner.", "sentence2": "we will study the problem in two different settings: when Learner is aware of this order and when it is not.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_251", "sentence1": "In [7], the comparator set is the set of all probability vectors over K experts, while we compare Learner's performance to the performance of the best ordering.", "sentence2": "the problem considered in [7] aims to compare Learner's performance to the best mixture of actions, which also includes our comparator set (orderings).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_252", "sentence1": "We also performed a qualitative comparison of the latent structures inferred by the different models and found that the sparse variant of the PRGDS inferred some components that the other models did not.", "sentence2": "the sparse variant of the PRGDS is uniquely capable of inferring bursty latent structures that are highly localized in time; we visualize examples in Fig. 5.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_253", "sentence1": "We recommend using these layers when generative modeling with normalizing flows (Dinh et al., 2017) or understanding how networks make predictions (Jacobsen et al., 2018).", "sentence2": "our layers tie the model specification to the inference algorithm (typically, variational inference).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_254", "sentence1": "We initialized a random matrix B ∈ Rk⇥m, testing different values for m", "sentence2": "we find that our trained probes are able to achieve slightly improved accuracy down to m = 128.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_255", "sentence1": "This dataset has the constituency grammar for the sentences, which was translated to a dependency grammar using the PyStanfordDependencies library [14].", "sentence2": "we clamped the cosine similarity terms to within ±0.1 of the pre-training averages for same and different senses.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_256", "sentence1": "If we do not assume Assumption 1, the arguments in Theorem 2 easily extend to showing instead that we approach some identified ✓ that satisfies all moment conditions.", "sentence2": "using a rich class of moment conditions allows us to learn correspondingly a rich g0 .", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_257", "sentence1": "In particular this means that if we parametrize f and g via neural networks where we can permute the parameter vector ✓ and obtain an identical function, our result still holds.", "sentence2": "as long as F has bounded complexity, even if its ambient dimension may be infinite, we can guarantee the consistency of our approach.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_258", "sentence1": "When g(x; θ) is a flexible model such as a high-capacity neural network, many -possibly infinitely many -moment conditions may be needed to identify θ 0 .", "sentence2": "gMM and OWGMM algorithms fail when we use too many moment conditions.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_259", "sentence1": "Instead, our approach is to leverage our variational reformulation in Lemma 1 and replace the class of functions F with a rich (non-subspace) set in this new formulation, which is distinct from GMM and avoids these issues.", "sentence2": "as long as F has bounded complexity, even if its ambient dimension may be infinite, we can guarantee the consistency of our approach.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_260", "sentence1": "On the other hand, we also cannot hope to learn the optimal weighting: the matrix Co in Eq. (6) will necessarily be singular and using its pseudoinverse would mean deleting all but n moment conditions.", "sentence2": "we cannot simply use infinite or even too many moment conditions in GMM or OWGMM.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_261", "sentence1": "Using first-order iterative algorithms for solving Eq. ( 9) enables us to effectively handle very large datasets.", "sentence2": "we implement DeepGMM using PyTorch, which efficiently provides gradients for use in our descent algorithms [30].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_262", "sentence1": "The rationale behind the clustering penalty was to better disentangle the features in the latent space.", "sentence2": "we obtain latent representations where the different clusters (i.e., classes) are better separated, thereby facilitating their modeling.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_263", "sentence1": "Alternatively, using a nonparametric bivariate copula estimator provides the required flexibility.", "sentence2": "the bivariate Gaussian kernel estimator, targeted at densities of unbounded support, cannot be directly applied to pair-copulas, which are supported in the unit square.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_264", "sentence1": "It should be noted that, in both cases, the nonparametric estimator captures the fact that X 1 and X 2 are independent, as can be seen from the contour densities on the left panel.", "sentence2": "adversarial approaches require multiple models to be trained, leading to difficulties and computational burden [62,26,24], and variational approaches make (strong) distributional assumptions, potentially detrimental to the generative model performance [64].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_265", "sentence1": "However, AEs simply learn the most informative features to minimize the reconstruction loss, and therefore cannot be considered as generative models.", "sentence2": "since they do not learn the distributional properties of the latent features [5], they cannot be used to sample new data points.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_266", "sentence1": "However, expected future rewards equal to zero are in general not possible for MDPs.", "sentence2": "we introduce sequence-Markov decision processes (SDPs), for which reward distributions need not to be Markov.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_267", "sentence1": "We can use any contribution analysis method, but we specifically consider three methods: (A) differences of return predictions, (B) integrated gradients (IG) [46], and (C) layer-wise relevance propagation (LRP) [3].", "sentence2": "this task has probabilistic state transitions, which can be represented as a tree with states as nodes.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_268", "sentence1": "FNNs are prone to such prediction errors since they have to predict the expected return again and again from each different state-action pair and cannot use stored information.", "sentence2": "the LSTM is less prone to produce spurious rewards: (i) The LSTM will only learn to store information if a state-action pair has a strong evidence for a change in the expected return.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_269", "sentence1": "The advantage function is the sales price minus the expected immediate repair costs minus the expected future delivery costs.", "sentence2": "you want to know whether the advantage function is positive.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_270", "sentence1": "Learning with non-optimal reward redistributions does work since the optimal policies do not change according to Theorem 1.", "sentence2": "reward redistributions that are optimal considerably speed up learning, since future expected rewards introduce biases in TD methods and high variances in MC methods.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_271", "sentence1": "For example, any contributions would be incorrect if the true expectation of the return did not change.", "sentence2": "prediction errors might falsely cause contributions leading to spurious rewards.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_272", "sentence1": "The reward can be viewed to be probabilistic but is prone to have high variance.", "sentence2": "we prefer method (A).", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_273", "sentence1": "The sum of the Q-value differences gives the difference between expected return at sequence begin and the expected return at sequence end (telescope sum).", "sentence2": "q-value differences allow to predict the expected return of the whole state-action sequence.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_274", "sentence1": "These tasks are designed to show problems of TD, MC, and potential-based reward shaping.", "sentence2": "reward redistributions that are optimal considerably speed up learning, since future expected rewards introduce biases in TD methods and high variances in MC methods.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_275", "sentence1": "Consecutive predictions share the same past and the same future contributions except for two immediate state-action pairs.", "sentence2": "in the difference of consecutive predictions contributions cancel except for the two immediate state-action pairs.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_276", "sentence1": "The goal is to learn that a delayed reward is larger than a distracting immediate reward.", "sentence2": "the correct expected future reward must be assigned to many state-action pairs.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_277", "sentence1": "According to Theorem 1, reward redistribution keeps the optimal policies.", "sentence2": "even nonoptimal reward redistributions ensure correct learning.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_278", "sentence1": "In contrast to standard empirical risk minimization in supervised learning, where test data follow the same distribution as training data, minimax statistical learning arises in problems of distributionally robust learning [16,18,28,29,40] and minimizes the worst-case risk over a family of probability distributions.", "sentence2": "it can be applied to the learning setting in which the test data distribution differs from that of the training data, such as in domain adaptation and transfer learning [12].", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_279", "sentence1": "The bound may be loose in some cases, since we consider the worst case distribution in the Wasserstein ball to avoid computing the transport map.", "sentence2": "for some problems, it may be possible to derive the transport map and thus get tighter bounds.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_280", "sentence1": "Machine learning models, especially deep neural networks, have achieved impressive performance across a variety of domains including image classification, natural language processing, and speech recognition.", "sentence2": "these techniques can easily be fooled by adversarial examples, i.e., carefully perturbed input samples aimed to cause misclassification during the test phase.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_281", "sentence1": "Then, Lee & Raginsky [28] derived the risk bound for minimax learning by exploiting the dual representation of worst-case risk.", "sentence2": "their minimax risk bound would go to infinity and thus become vacuous as eb-> 0.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_282", "sentence1": "Finally we note that the shape of M remains similar to tanh(E).", "sentence2": "secondly, we introduce a secondary de-attention (deleted attention) matrix, finally learning a multiplicative composition of similarity and dissimilarity.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_283", "sentence1": "(2) As a result, while this robust prediction objective falls within our framework, and regularizes robust attributions, it allows a small regularization term where attributions actually change significantly but they cancel each other in summation.", "sentence2": "the control over robust attributions can be weak.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_284", "sentence1": "Our theory is grounded in the recent work, Integrated Gradients (IG) [STY17], in axiomatically attributing a neural network's output change to its input change.", "sentence2": "an emerging problem to tackle in this domain is to train models that produce reliable interpretations for their predictions.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_285", "sentence1": "While, in general, f H may be difficult to compute, we show settings in which it is tractable.", "sentence2": "for kernel ridge regression with a Gaussian kernel, we prove a bound on f H that, as a byproduct, yields generalization bounds that match (up to a small constant) the standard ones.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_286", "sentence1": "To circumvent this issue, next we approach the DRO problem from a different angle: we directly search for the adversarial distribution Q.", "sentence2": "for example, f and f may not belong to the same RKHS -it is not hard to construct counterexamples, even when is merely quadratic.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_287", "sentence1": "External memory may be helpful or even necessary for agents operating in partially observable environments, where current decision making depends on a sequence of past observations.", "sentence2": "memory mechanisms also add complexity to an agent's learning process [19], since the agent must learn to access its memory during experience.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_288", "sentence1": "Because these feedback connections are not computed nor used sequentially, updates can be parallelized for speed.", "sentence2": "fixed random feedback connections may be inefficient.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_289", "sentence1": "The optimal parameter ter can be obtained by maximizing the lower bound in (2), which are for the datapoints of seen classes.", "sentence2": "by sampling the unseen datapoints while reducing the model uncertainty, the model better describes the target distribution of unseen classes.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_290", "sentence1": "To relieve the problem, SGAL-dropout uses dropout activation when sampling unseen datapoints and shows more robust classification results compared to that of SGAL's.", "sentence2": "by sampling the unseen datapoints while reducing the model uncertainty, the model better describes the target distribution of unseen classes.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_291", "sentence1": "The combination of the large amount of data and deep learning finds the usage in various fields such as machine learning and artificial intelligence.", "sentence2": "deep learning as a non-linear regression tool based on statistics mostly suffers from the insufficient or non-existing training data, which is the usual case and should be overcome for autonomous learning systems.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_292", "sentence1": "This optimal generative model can be obtained with (5) by trained on that sampled datapoints X u * of unseen classes, and existing datapoints of seen classes.", "sentence2": "we can have a generative model which covers both seen and unseen classes by obtaining optimal parameters and sampled datapoints that satisfy (4) and (5).", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_293", "sentence1": "In order to circumvent the need for samples of unseen classes, we treat the non-existing data as missing examples.", "sentence2": "our network aims to find optimal unseen datapoints and model parameters, by iteratively following the generating and learning strategy.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_294", "sentence1": "In other words, while training our model iteratively generate unseen samples and use them as training datapoints to gradually update model parameters.", "sentence2": "our model favorably attain both seen and unseen classes understanding.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_295", "sentence1": "Compared to the Gumbel-Softmax estimator, our approach relies on maximization over the latent variable assignments, rather than summation, which is computationally more efficient.", "sentence2": "performing maximization exactly or approximately is possible in many structured cases, even when summation remains intractable.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_296", "sentence1": "The typical solution is to replace the desired objective with a surrogate differentiable loss, such as the cross-entropy loss between the targets and the predicted distribution over labels.", "sentence2": "the direct loss minimization approach proposes to minimize the desired objective directly.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_297", "sentence1": "One can also see that the general structured encoder, with any αi,j (x), achieves better test loss than the super-modular structured encoder.", "sentence2": "this comes with a computational price, as the maxflow algorithm is orders of magnitude faster than CPLEX, and structured encoder with CPLEX becomes better than maxflow only in epoch 85, see Figure 3.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_298", "sentence1": "[12] use the Gumbel-Max trick to reparameterize the discrete VAE objective, but, unlike our work, they relax the resulting formulation, replacing the arg max with a softmax operation.", "sentence2": "they introduce the continuous Concrete (Gumbel-Softmax) distribution and replace the discrete random variables with continuous ones.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_299", "sentence1": "The Perturb-and-Parse approach [5] focuses on latent dependency parses, and iteratively replaces any arg max with a softmax operation in a spanning tree algorithm.", "sentence2": "our framework is not restricted to a particular class of structures.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_300", "sentence1": "This unsatisfactory pattern of quantization error in Theorem 3 comes from the finite sample setting and proof methodology, since the bound is a worst case bound with n and k both finite.", "sentence2": "this bound is less meaningful for practical purposes.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_301", "sentence1": "At this point, 1-bit LM quantizer has much smaller debiased variance than others.", "sentence2": "we expect 1-bit LM to provide highest test accuracy on these two datasets, which is again consistent with Figure 4.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_302", "sentence1": "Throughout this paper, we assume that every sample in X is standardized to have unit Euclidean norm 2 .", "sentence2": "the domain of X is the unit Euclidean sphere S d , which allows us to call \"inner product\" and \"cosine\" interchangeably.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_303", "sentence1": "Hence, for 1-NN classification, we should ideally choose quantizers with low debiased variance around p * = cos(x, x i ), provided that it can be known (or estimated) a priori.", "sentence2": "quantizers with small debiased variance are favorable for NN classifier and linear classifier, in high similarity region and around ρ = 0, respectively.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_304", "sentence1": "Here, we look at this problem in the asymptotic domain.", "sentence2": "intuitively, label flipping is much more likely to occur for the points near the boundary (i.e., with small ĥT x).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_305", "sentence1": "These results provide an encouraging case for disentanglement being helpful in finding fairer representations.", "sentence2": "roadmap: In Section 2, we briefly review the state-of-the-art approaches to extract and evaluate disentangled representations.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_306", "sentence1": "As one builds machine learning models for different tasks on top of such general purpose representations, it is not clear how the properties of the representations relate to the fairness of the predictions.", "sentence2": "for different downstream prediction tasks, there may be different sensitive variables that we would like to be fair to.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_307", "sentence1": "As a sanity check, we finally confirm by visual inspection that the adjusted metrics still measure disentanglement.", "sentence2": "we believe that these results serve as a motivation for further investigation on the practical benefits of disentangled representations, especially in the context of fairness.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_308", "sentence1": "With this we compute the disentanglement metrics and use the following score to measure the unfairness of the predictions where T V is the total variation.", "sentence2": "we compare the average total variation of the prediction after intervening on s, thus directly measuring the violation of demographic parity.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_309", "sentence1": "We clearly observe that learned representations can be unfair, even in the setting where the target variable and the sensitive variable are independent.", "sentence2": "the total variation can reach as much as 15% 25% on five out of seven data sets.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_310", "sentence1": "We remark that this setup is different from what we consider in this paper, as we do not assume access to any labeled information when learning a representation.", "sentence2": "we do not assume to know what the downstream task will be and what are the sensible variables (if any).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_311", "sentence1": "If we limit the memory usage, then this implies that the effective capacity of the replay is limited as any transitions we replace are forgotten completely.", "sentence2": "parametric models may be able to achieve good accuracy with a fixed and comparatively small memory footprint.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_312", "sentence1": "Sometimes, there may be no practical differences between replay and models, depending on how they are used.", "sentence2": "a replay memory is less flexible than a model, since we cannot query it at arbitrary states that are not present in the replay memory.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_313", "sentence1": "Although the distinction is not fully unambiguous, there exist two prototypical families of algorithms: those that learn without an explicit model of the environment (model free), and those that first learn a model and then use it to plan a solution (model based).", "sentence2": "this has implications for Dyna-style learning, as well as for replay methods [cf.. van Hasselt et al., 2018].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_314", "sentence1": "Instead, planning backward when the model is inaccurate may lead to updates to fictional states that are unreachable, which may or may not be useful, but is less likely to be harmful than updating real states with fictional transitions.", "sentence2": "if, however, we sample states according to a non-uniform distribution (e.g., using prioritised replay) this can make replay-based algorithms less stable and potentially divergent [cf.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_315", "sentence1": "Experience replay [Lin, 1992] refers to storing previously observed transitions to replay later for additional updates to the predictions and policy.", "sentence2": "one of the main results by Kaiser et al. [2019] was to compare SimPLe to Rainbow DQN [Hessel et al., 2018a], which combines the DQN algorithm [Mnih et al., 2013[Mnih et al., , 2015 with double Q-learning [van Hasselt, 2010, dueling network architectures [Wang et al., 2016], prioritised experience replay , noisy networks for exploration [Fortunato et al., 2017], and distributional reinforcement learning [Bellemare et al., 2017].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_316", "sentence1": "We examine the question of when and how parametric models are most useful in reinforcement learning.", "sentence2": "we look at commonalities and differences between parametric models and experience replay.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_317", "sentence1": "Both of these methods can only work on feed-forward networks that are separable between layers.", "sentence2": "neither approach can parallelize Transformer-based language models, because the shared embeddings make the networks non-separable.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_318", "sentence1": "Empirical results demonstrate the superiority of Transformer networks and show that a larger model tends to yield better performance.", "sentence2": "in the next section, we analyze the convergence rate of Algorithm 1, which is the basis of analysis for other variants of SGD.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_319", "sentence1": "It is an ideal case to obtain linear speedup, using K × machines to achieve K × speedup regarding time.", "sentence2": "it is impossible to achieve even for data parallelism.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_320", "sentence1": "In the proposed algorithm, the backward computation in module k is always one time step behind module k + 1.", "sentence2": "the computations in all modules can be parallelized.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_321", "sentence1": "We split an L-layer network into K modules so that the weights of the network are divided into K groups and each group is placed on a GPU.", "sentence2": "we have w = [w G (1) , w G(2) , ..., w G(K) ] where G(k) denotes layer indices in group k. ", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_322", "sentence1": "Empirical results demonstrate the superiority of Transformer networks and show that a larger model tends to yield better performance.", "sentence2": "when a model is so large that it has to be allocated on multiple GPUs, data parallelism over these GPUs is not applicable because it requires each GPU to have one copy of the whole model.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_323", "sentence1": "Consequently, we propose to store the input of each module as well as a random seed.", "sentence2": "before computing activations, we initialize the random number generator in GPU with the stored seed.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_324", "sentence1": "Recent efforts in deep learning models interpretation allow us to understand the decision for particular examples Ribeiro et al. (26); Shrikumar et al. (27); Štrumbelj, Kononenko (31); Koh, Liang (15); Lundberg, Lee (19), but understanding the logic behind a complex model is still a challenging task.", "sentence2": "symmetric trees outperform other tree shapes: they are coded by a decision table, thus evaluation for one tree requires several bit-wise operations and one look-up in the table and takes >10x less time for the same number of trees Dorogush et al. (6).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_325", "sentence1": "There is no universal strategy for tree induction: optimal one changes with dataset at hand.", "sentence2": "symmetric trees outperform other tree shapes: they are coded by a decision table, thus evaluation for one tree requires several bit-wise operations and one look-up in the table and takes >10x less time for the same number of trees Dorogush et al.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_326", "sentence1": "CatBoost (25) uses a level-wise policy, similar to one used in XGBoost, but with one more restriction: CatBoost searches for just one single condition to split all leaves simultaneously.", "sentence2": "the result is, effectively, a decision table.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_327", "sentence1": "This set optimization is clearly NP-hard: there is a set cover problem under the shell 3 .", "sentence2": "the target of the optimization is submodular, so it is possible to find an approximately optimal solution in polynomial time Bach (1).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_328", "sentence1": "One desired property of such a probabilistic model is sharpness (or high accuracy), i.e., if possible, the model should assign the highest probability to the true tumour grade (which maybe can not be inferred from the image at hand but only by other means such as an additional immunohistochemical staining).", "sentence2": "to be able to trust the predictions the probabilities should be calibrated (or reliable) as well (DeGroot and Fienberg, 1983;Murphy and Winkler, 1977).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_329", "sentence1": "While in tree-based methods, the tree index's hierarchy also affects the retrieval model training.", "sentence2": "how to learn the tree index and user preference model jointly is an important problem.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_330", "sentence1": "A series of related work [30,6,18,16,32,33,34] has shown that the user's historical behaviors play a key role in predicting the user's interests.", "sentence2": "the corpus size, and there is no restriction on the preference model structure.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_331", "sentence1": "This further decomposition is important in understanding how the ensemble model's predictive uncertainty changes by accounting for the fact that its prediction and distribution functions may be misspecified.", "sentence2": "recall that an ensemble model's parametric uncertainty is the uncertainty about the ensemble weights under the current model specification (i.e. by assuming δ = 0,G = I).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_332", "sentence1": "We vary sample size between 100 and 1000, and repeat the simulation 50 times in each setting.", "sentence2": "bAE's aleatoric uncertainty is still biased in that it is roughly constant throughout the range of x, failing to account for the heterogeneity in observation's variance, a pattern that is evident in data's empirical distribution.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_333", "sentence1": "As shown in Figure 1, the posterior uncertainty in each of a BNE's model parameters {ω,δ,G} accounts for an important source of model uncertainty.", "sentence2": "both the aleatoric and epistemic uncertainties are quantified by the BNE's posterior distribution, and can be distinguished through a careful decomposition of the model posterior.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_334", "sentence1": "Aleatoric uncertainty arises due to the stochastic variability inherent in the data generating process, for example due to an imperfect sensor, and is described by the cumulative distribution function (CDF) F((y|x,Θ) of the data specified by a given model.", "sentence2": "epistemic uncertainty arises due to our lack of knowledge about the data generating mechanism.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_335", "sentence1": "In practice, semi-supervised learning methods with many hyperparameters can be problematic because cross-validation is difficult with small validation sets [35,39,35].", "sentence2": "we find in practice that most of MixMatch's hyperparameters can be fixed and do not need to be tuned on a per-experiment or per-dataset basis.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_336", "sentence1": "BCOP has all of its singular values at 1 throughout training by design due to its gradient norm preservation and orthogonality.", "sentence2": "we empirically verify the downsides of other methods and show that our proposed method enables maximally expressive Lipschitz constrained convolutional layers with guaranteed gradient-norm-preservation.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_337", "sentence1": "The property above (Equation 2) allows one to upper-bound the Lipschitz constant of a network by the product of the Lipschitz constants of each layer.", "sentence2": "as modern neural networks tend to possess many layers, the resultant upper-bound is likely to be very loose, and constraining it increases the risk of diminishing the Lipschitz constrained network capacity that can be utilized.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_338", "sentence1": "The Lipschitz constant (or spectral norm) of a convolution operator is bounded by a constant factor of the spectral norm of its reshaped matrix [12,48,40], which enables bounding of the convolution operator's Lipschitz constant by bounding that of the reshaped matrix.", "sentence2": "this upper-bound can be conservative, causing a bias towards convolution operators with low Lipschitz constants, limiting the method's expressive power.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_339", "sentence1": "We also evaluate the empirical robustness of our model around under two gradient-based attacks and two decision-based attacks: (i) PGD attack with CW loss [34,7], (ii) FGSM [47], (iii) Boundary attack (BA) [6], (iv) Point-wise attack (PA) [44].", "sentence2": "the gradient-based methods ((i) and (ii)) are done on the whole test dataset; the decision-based attacks ((iii) and (iv)) are done only on the first 100 test data points since they are expensive to run.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_340", "sentence1": "How to modify the training data with bounded transferable perturbation that can lead to the largest generalization gap?", "sentence2": "we consider the task of adding imperceivable noises to the training data, hoping to maximally confuse any corresponding classifier so as to make wrong predictions as much as possible when facing clean test data.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_341", "sentence1": "For instance, the attacker wants the classifier to wrongly recognize the pattern from class A specifically to Class B (thus not to Class C).", "sentence2": "by teaching the perturbation generator to hijacking the training trajectory of the victim classifier, the generator can thus learn to move against the victim classifier step by step.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_342", "sentence1": "Mathematically, in the case of transformation sets, the latter can be modeled by a property of the true distribution.", "sentence2": "[39,26], the S-PGD mechanism uses projected gradient descent with respect to the translation and rotation parameters with projection on the constrained set S of transformations.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_343", "sentence1": "Out of curiosity we also trained a \"two-stage\" STN where we train the localization network separately in a supervised fashion.", "sentence2": "we use a randomly transformed version of the training data, treating the transformation parameters as prediction targets.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_344", "sentence1": "Unfortunately, the relationship between power and these parameters changes with the underlying distribution of the null and alternate p-values, as well as their relative frequency.", "sentence2": "below, we only provide a heuristic argument about how to tune these parameters for ADDIS * .", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_345", "sentence1": "In this section, we numerically compare the performance of ADDIS against the previous state-ofthe-art algorithm SAFFRON [5], and other well-studied algorithms like LORD++ [4], LOND [10] and Alpha-investing [2].", "sentence2": "we use ADDIS * defined in Algorithm 1 as the representative of our ADDIS algorithm.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_346", "sentence1": "In Section 2, we derive the ADDIS algorithm and state its guarantees (FDR and mFDR control), deferring proofs to the supplement.", "sentence2": "in Section 2.4, we discuss how to choose the hyperparameters in ADDIS to balance adaptivity and discarding for optimal power.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_347", "sentence1": "ADDIS is based on a new serial estimate of the false discovery proportion, having adaptivity to both fraction of nulls (like SAFFRON) and the conservativeness of nulls (unlike SAFFRON).", "sentence2": "we present the following generalizations and leave the details (formal setup, proofs) to supplement for interested readers.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_348", "sentence1": "Doing the computation in an ASIC would drastically reduce the latency of kinematics inference and eliminate a large power draw for the gigabytes of neural data that must be transferred otherwise.", "sentence2": "we plan to create an ASIC that can be implanted in the brain to perform inference of kinematics from neural signals.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_349", "sentence1": "Moreover, because of state feedback, a small recurrent network can be equivalent to a large feed-forward network.", "sentence2": "a recurrent network will be computationally efficient, especially for the applications that require hardware implementation [19].", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_350", "sentence1": "During inference, since the ground truth values are unavailable, the feedback, z k , has to be replaced by the previous network predictions.", "sentence2": "the same approach cannot be applied during training since the DRNN has not been trained yet and it may cause poor performance of the DRNN.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_351", "sentence1": "Because of this disparity between training and testing, the DRNN may enter unseen regions of the state-space, leading to mistakes at the beginning of the sequence prediction process.", "sentence2": "we should find a strategy to start from the ground truth distribution and move toward the predictions' distribution slowly as the DRNN learns.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_352", "sentence1": "These characteristics of PPC differentiate it from other brain areas and, while providing a large amount of information to the decoder, also require new paradigms, such as those discussed here, to extract useful information.", "sentence2": "extracting appropriate neural features and designing a robust decoder that can model this relationship in an actual BMI setting is required.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_353", "sentence1": "Our results are shown in Table 2.", "sentence2": "6-HALF MOONS interpolates the original half moons dataset using 6 points with added noise.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_354", "sentence1": "This means one needs to store O(#nodes X #message passing steps) states, which can be costly for large graphs.", "sentence2": "gRevNets can reconstruct hidden states in lower layers from higher layers during backpropagation, meaning one only needs to store O(#nodes) states.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_355", "sentence1": "Machine learning has been quite successful at generative modeling of complex domains such as images, audio, and text.", "sentence2": "relational data poses new and interesting challenges such as permutation invariance, as permuting the nodes results in the same underlying graph.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_356", "sentence1": "Our GNF consists of 10 MP steps with attention and an MLP for each of F 1 , F 2 , G 1 , and G 2 .", "sentence2": "this poses a problem for certain aggregation functions like sum and mean.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_357", "sentence1": "This allows the model to learn how to organize nodes in order to match a specific distribution.", "sentence2": "this poses a problem for certain aggregation functions like sum and mean.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_358", "sentence1": "The result of using a fully connected graph is that the computational cost of message passing is O(N 2 ), similar to the GraphRNN.", "sentence2": "each step of the GNF is expressible in terms of matrix operations, making it more amenable to parallel architectures.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_359", "sentence1": "The initial values for v i (0) and u i (0) are set to -65 and -13 respectively for all nodes.", "sentence2": "biological neural networks in the brain grow architecture that can generalize very well to innumerable datasets.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_360", "sentence1": "Among all the short sequences, we randomly choose 100,000 sequences for each class for training, validation, and test.", "sentence2": "we feed the one-hot encoded DNA sequences into an LSTM layer, followed by a dense layer and a softmax function to predict the probability distribution over the 4 letters of {A, C, G, T }, and train the model using only the in-distribution training data.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_361", "sentence1": "Please see supplementary for the algorithms and time complexities of all the proposed methods and HGNN.", "sentence2": "a simple yet effective way to overcome the limitations is to introduce hyperedge-dependent vertex weights [14].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_362", "sentence1": "They study sketching algorithms from both statistical and algorithmic perspectives.", "sentence2": "they focus on a different setting, where n >>\u001d r, and prove bounds on RE and P E. For instance, they discover that RE can be bounded even when r is not too large, proving bounds such as RE ≤ 1 + 44p/r for subsampling and subgaussian projections", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_363", "sentence1": "(Drineas et al., 2012), show furthermore that leverage scores can be approximated fast using the Hadamard transform.", "sentence2": "suppose also that X is a deterministic matrix s.t. the esd. of XTX converges weakly to some fixed probability distribution with compact support bounded away from the origin.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_364", "sentence1": "Various versions of this fundamental problem have been studied before (see e.g., Drineas et al., 2006Drineas et al., , 2011Dhillon et al., 2013;Ma et al., 2015;Raskutti and Mahoney, 2016;Thanei et al., 2017, and the references therein).", "sentence2": "in a generative model where the data are sampled from a linear regression model, Raskutti and Mahoney (2016) have recently compared the statistical performance of various sketching algorithms, such as Gaussian projections and subsampled randomized Hadamard transforms (SRHT) (introduced earlier in Sarlos (2006); Ailon and Chazelle (2006)).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_365", "sentence1": "Various versions of this fundamental problem have been studied before (see e.g., Drineas et al., 2006Drineas et al., , 2011Dhillon et al., 2013;Ma et al., 2015;Raskutti and Mahoney, 2016;Thanei et al., 2017, and the references therein).", "sentence2": "this is because iid projections distort the geometry of Euclidean space due to their non-orthogonality.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_366", "sentence1": "We observe that our results are accurate, both in simulations and in two empirical data analysis examples, see Section 3.", "sentence2": "they go beyond earlier work (Raskutti and Mahoney, 2016) because they are accurate not just up to the rate, but also down to the precise constants, even in relatively small samples (see Section A.16 in the supplemental for a comparison).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_367", "sentence1": "Fast orthogonal transforms such as the Hadamard transforms are considered as a baseline for sketching methods, because they are efficient and work well quite generally.", "sentence2": "if the data are very uniform, for instance if the data matrix can be assumed to be nearly rotationally invariant, then sampling methods can work just as well, as will be shown below.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_368", "sentence1": "To encode this form of equivariance for human pose regression tasks, we propose \"chirality nets.\"", "sentence2": "the output of a chirality net is guaranteed to be equivariant w.r.t. a transformation composed of reflections and label switching. ", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_369", "sentence1": "For instance, collaborative virtual reality applications rely on accurate pose estimation for which significant advances have been reported in recent years.", "sentence2": "recent state-of-the-art approaches use supervised learning to address pose regression and employ deep nets.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_370", "sentence1": "For instance, collaborative virtual reality applications rely on accurate pose estimation for which significant advances have been reported in recent years.", "sentence2": "additionally, when comparing without test-time augmentation, our approach outperforms by 1 mm.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_371", "sentence1": "Pose symmetric 1D convolution layers can be based on fully connected layers.", "sentence2": "a 1D convolution is a fully connected layer with shared parameters across the time dimension, i.e., at each time step the computation is the sum of fully connected layers over a window: we enforce equivariance at each time step by employing the symmetry pattern of fully connected layers at each time slice.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_372", "sentence1": "The temporal convolution considers the intra-frame information while the spatial convolution considers the inter-frame information.", "sentence2": "recent state-of-the-art approaches use supervised learning to address pose regression and employ deep nets.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_373", "sentence1": "If both gate and cell preserve the negation then the product will not.", "sentence2": "we change the weight sharing scheme for the gates.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_374", "sentence1": "The proposed model, which achieves compelling results, leverages those cues by combining a long-short-term-memory (LSTM) module based deep net with attention over objects to obtain grounding and context.", "sentence2": "the proposed model is also very intricate.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_375", "sentence1": "Moreover, notice that the elements of diagonal confusions (d's) and off-diagonal confusions (c's) reflect correct and incorrect classification, respectively.", "sentence2": "according to standard practice, wlog., we focus on eliciting monotonically increasing DLPMs and monotonically decreasing LPMs in their respective arguments.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_376", "sentence1": "For any query based approach, it is important to understand the structure of the query space.", "sentence2": "we first study the properties of the query spaces and then develop parametrizations required for efficient elicitation.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_377", "sentence1": "Instead, we provide novel strategies for eliciting linear functions of the multiclass confusion matrix and extend elicitation to more complicated yet popular functional forms such as linear-fractional functions of the confusion matrix elements [14].", "sentence2": "the elicitation procedures involve binary-search type algorithms that are robust to both finite sample and oracle feedback noise.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_378", "sentence1": "Finite samples may affect the size of the sphere Sλ in LPM elicitation, but we observe that as long as λ is greater than \u000fΩ LPMs can be elicited (Appendix F.2).", "sentence2": "here we emprically validate only DLPM elicitation with finite samples.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_379", "sentence1": "Metric Elicitation is a principled framework for selecting the performance metric that best reflects implicit user preferences.", "sentence2": "available strategies have so far been limited to binary classification.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_380", "sentence1": "By using Lemma 1, obtain its BO off-diagonal confusions over the sphere Sλ , which clearly lies on the lower boundary.", "sentence2": "varying θ, in this procedure, parametrizes the lower boundary ∂Sλ .", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_381", "sentence1": "The term z 1 log(z 2 /(q 2 )) may attribute to the number of cycles in Algorithm 2, but due to the curvature of the sphere, we observe that it is not a dominating factor in the query complexity.", "sentence2": "intuitively, this weak assumption ensures that when the cost or reward tradeoffs for the classes change, the preferred confusion matrices for those cost or reward tradeoffs also change (and vice-versa).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_382", "sentence1": "We propose to assess bias at the contextual word level.", "sentence2": "this is true of all instances of significant effect sizes on test +I5 except for the c-word encoding of BERT (bbc).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_383", "sentence1": "This seems to suggest a degree of anti-stereotypical associations, but none of the tests which have negative effect sizes are significant Table 5: Intersectional embedding association tests and effect sizes, for word encodings (word), sentence encoding (sent) and contextual word representation (c-word).", "sentence2": "there are only 2 significant positive associations for test +I4, but there are 9 such associations for test +I3 across all models.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_384", "sentence1": "Since the WebText dataset has not been released fully by [27], we only use the partially released version with 250K documents 2 .", "sentence2": "barring a more robust study with more model sizes in consideration, we caution against a definitive conclusion.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_385", "sentence1": "We depart from previous work by including the nominative (she), accusative (her), prenominal possessive (her) and predicative possessive (hers) inflections of personal pronouns, and we also include the non-gendered or collective pronoun they.", "sentence2": "for each sentence, we increment a count for female pronoun occurrence if any female pronoun is in the sentence, and then count the number pro-stereotypical and anti-stereotypical associations with occupation words (same for male and non-gendered pronouns).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_386", "sentence1": "Although these methods are promising, Gonen and Goldberg [13] show that the gender bias encoded in word embeddings is mostly hidden from the defined metric of projecting onto the \"he-she\" vector direction.", "sentence2": "words still receive implicit gender from their associations (e.g. \"receptionist\" is no longer gendered with respect to \"he\", but is still gendered with respect to \"captain\"). ", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_387", "sentence1": "Moreover, we find stronger evidence for bias when comparing against race (test +I3) than when comparing against gender (test +I4).", "sentence2": "there are only 2 significant positive associations for test +I4, but there are 9 such associations for test +I3 across all models.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_388", "sentence1": "This indicates that when assessing social bias, the type of encoding matters determines whether the bias can be measured.", "sentence2": "contextual word representations can be used in conjunction with sentence encodings to determine bias in a given model.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_389", "sentence1": "Fifth, we introduce a method of comparison that anchors at the most or least privileged group to show that intersectional identities suffer from such bias as well, and more so than their minority identities.", "sentence2": "we show that the effect of race on intersectional identities seems to be larger than the effect of gender.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_390", "sentence1": "Emerging Convolutions use \"causal masks\"  whose inverse function falls into a complete autoregressive transformation.", "sentence2": "mACOW achieves significantly more efficient inference and sampling ( §4.3), due to the carefully designed masks (Figure 1).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_391", "sentence1": "Compared with Theorem 2, the lower bounds in Theorem 3 lose a polynomial factor in M due to a larger policy space.", "sentence2": "since the number of batches M of interest is at most O(log T ) (otherwise by Corollary 1 we effectively arrive at the fully adaptive case with M = T ), this penalty is at most poly-logarithmic in T .", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_392", "sentence1": "In the case of variable noise, our method remains roughly as good when the noise parameters are known, but starts to have trouble when they need to be estimated from data.", "sentence2": "with additive Gaussian noise the product involves two Gaussians, and because both distributions are functions of x, we have where we have exploited the symmetry of Gaussian distribution in the first term to swap x and y.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_393", "sentence1": "For example, at x-axis position marked \"1\" they have been trained for 2M minibatches compared to 0.5M minibatches for our method.", "sentence2": "the receptive field of the resulting network includes the center pixel, so we offset the feature maps by one pixel before combining them.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_394", "sentence1": "Unfortunately, this does not imply that our self-supervised technique would be guaranteed to find the same optimum: we approximate the prior distribution with a Gaussian, whereas standard supervised training corresponds to a Gaussian approximation of the posterior.", "sentence2": "a single-pixel offset at the end of each branch separates the receptive field from the center pixel.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_395", "sentence1": "The approximate distribution p(x|Ω y ) allows us to now apply Bayesian reasoning to include information from y at test-time.", "sentence2": "the (unnormalized) posterior probability of the clean value x given observations of both the noisy pixel y and its context is given by Bayes' rule as follows: From this point of view, the distribution p(x|Ω y ) takes the role of the prior, encoding our beliefs on the possible xs before observing y.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_396", "sentence1": "They are endowed with an LSH family via the Hellinger approximation.", "sentence2": "let P and Q be two different distributions on Ω.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_397", "sentence1": "Other examples are time series distributions, content of documents, or images that can be represented as histograms.", "sentence2": "analysis of similarities in time series distributions or documents can be used in the context of attacks, and spam detection.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_398", "sentence1": "All the intermediate activations were then \"re-computed\" by solving the ODE layers backwards.", "sentence2": "it has been recently shown that such an approach could lead to incorrect gradients, due both to numerical instability and also to inconsistencies that relate to optimizing infinite dimensional operators (the so called Discretize-Then-Optimize vs. Optimize-Then-Discretize issue) [9].", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_399", "sentence1": "In configuration one, a residual block is created for each of the three time steps.", "sentence2": "in configuration two, we only apply the first and last time evolutions of the parameters (i.e., we only use w 0 and w 1 to apply to activations).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_400", "sentence1": "We derive the optimality conditions for how backpropagation should be performed for the coupled ODE formulation using by imposing the standard Karush-Kuhn-Tucker conditions.", "sentence2": "we implement the corresponding Discretize-Then-Optimize (DTO) approach, along with a checkpointing scheme presented in [9].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_401", "sentence1": "The governing equation here is dz/dt = c(t) dz/dx, where c(t) is variable velocity, and z is the signal that changes in time.", "sentence2": "as discussed above, an alternative view of a residual network is the following continuous-time formulation: dz dt = f (z(t); ✓), with z(t = 0) = z 0 and z(t = 1) = z 1 (we will use both z(t) and z t to denote activation at time t).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_402", "sentence1": "Here we perform an ablation study in which we remove the evolution of the model parameters, and instead we fix them to stale values in time (which is the configuration used in [8,9]), and test with a case where the model parameters are indeed evolved in time, which corresponds to results of Table 2.", "sentence2": "we use two time steps for activation ODE (Eq. . 5b) and ten time steps for the evolution of the model parameters (Eq. 5c).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_403", "sentence1": "On the other hand, truly Bayesian inference would instead put a posterior distribution over model weights to characterize the uncertainty during training [2,20,30].", "sentence2": "due to the complexity of nonlinear neural networks, analytic posterior is not available, hence strong independent assumptions over model weight have to be made in order to achieve computationally tractable variational solution.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_404", "sentence1": "While most previous works aim to diversify the representations, we explore the complementary direction by performing an adaptive and data-dependent regularization motivated by the empirical Bayes method.", "sentence2": "we propose to construct a matrix-variate normal prior (on weights) whose covariance matrix has a Kronecker product structure.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_405", "sentence1": "Neural codes were shown to outperform existing state-of-the-art codes on the feedback channel [31].", "sentence2": "equipping a decoder with a learnable neural network also allows fast adaptation via meta-learning [25].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_406", "sentence1": "The encoder and the decoder together can be naturally viewed as an over-complete autoencoder, where the noisy channel in the middle corrupts the hidden representation (codeword).", "sentence2": "designing a reliable autoencoder can have a strong bearing on alternative ways of designing new encoding and decoding schemes for wireless communication systems.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_407", "sentence1": "There are two novel components of Turbo code which led to its success: an interleaved encoder and an iterative decoder.", "sentence2": "in the canonical setting of AWGN channel, neural codes are still far from capacity-approaching performance due to the following challenges.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_408", "sentence1": "The population size |P| is 50 and the parents size |P| is 10.", "sentence2": "there are 1000 networks evaluated in one search.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_409", "sentence1": "It can perform with a pre-trained backbone network and search with the previous NAS algorithm [31].", "sentence2": "table 4 shows the computation cost for each steps.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_410", "sentence1": "To this end, DetNAS chooses the evolutionary search algorithm.", "sentence2": "the search method and search space follow DetNAS.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_411", "sentence1": "The issue is that the batch statistics on one path should be independent of others.", "sentence2": "we need to recompute batch statistics for each single path (child networks) before each evaluation.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_412", "sentence1": "However, in supernet pre-training, we adopt a path-wise [7] manner to ensure the trained supernet can reflect the relative performance of candidate networks.", "sentence2": "in each iteration, only one single path is sampled for feedforward and backward propagation.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_413", "sentence1": "ShuffleNetv2-40 is inferior to ResNet-101 and DetNAS by 0.8% mmAP on COCO.", "sentence2": "for RetinaNet, the training settings is similar to FPN, except that the initial learning rate is 0.01.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_414", "sentence1": "Comparisons to the network for ImageNet classification.", "sentence2": "many object detectors directly use networks designed for Image classification as backbones.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_415", "sentence1": "We have shown that efficient coding models nonlinearly scale their resource allocation in sensory bottlenecks under nonuniform input densities.", "sentence2": "the limited number of outputs neurons is not simply allocated proportional to receptor density.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_416", "sentence1": "Eigenvalues chosen from region H imply that the receptive field of the added output neuron also falls onto region H 1 .", "sentence2": "the problem of how output neurons are allocated to either input region is solved by calculating and sorting the eigenvalues associated with each individual input region.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_417", "sentence1": "We show that different interactions between PMI vectors reflect semantic word relationships, such as similarity and paraphrasing, that are encoded in low dimensional word embeddings under a suitable projection, theoretically explaining why embeddings of W2V and GloVe work.", "sentence2": "we also reveal an interesting mathematical interconnection between the considered semantic relationships themselves.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_418", "sentence1": "Interestingly, the difference between the associated PMI vectors: is a vector of un-weighted KL divergence components.", "sentence2": "if dimensions were suitably weighted, the sum of difference components (comparable to Manhattan distance but directed) would equate to a KL divergence between induced distributions.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_419", "sentence1": "Whilst W2V and GloVe train two embedding matrices, typically only W is used and C discarded.", "sentence2": "although relationships are learned between W and C, they are tested between W and W. If W and C are equal, the distinction falls away, but that is not found to be the case in practice.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_420", "sentence1": "Hashimoto et al. [14] and Arora et al. [4] propose generative language models to explain the structure found in word embeddings.", "sentence2": "both contain strong a priori assumptions of an underlying geometry that we do not require (further, we find that several assumptions of [4] fail in practice (Appendix D)).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_421", "sentence1": "Intuitively, we can check if our alignment is correct via these reference measurements.", "sentence2": "we can obtain aligned, denoised inner products with each of the two parameter vectors.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_422", "sentence1": "Instead, we use a new method motivated by error-correcting codes.", "sentence2": "we perform several redundant queries, that help us to do this alignment.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_423", "sentence1": "Since there are L different vectors in the mixture, with O(L log L) measurements with a row we will be able to see the samples corresponding to each of the L vectors with that row.", "sentence2": "even if this is true for measurements with each rows, we will still not be able to align measurements across the rows.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_424", "sentence1": "For ReLU networks (piece-wise linear networks in general), exact verifiers solve the robustness verification problem (1) by typically employing MILP solvers [Cheng et al., 2017, Lomuscio and Maganti, 2017, Dutta et al., 2018, Fischetti and Jo, 2017 or Satisfiability Modulo Theories (SMT) solvers [Scheibler et al., 2015, Carlini et al., 2017, Ehlers, 2017.", "sentence2": "due to the NP-completeness for solving such a problem , it can be really challenging to scale these to large networks.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_425", "sentence1": "We do this by setting in (C) and computing the exact optimum.", "sentence2": "yet this improvement is not significant enough to close the gap with the lower bounds.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_426", "sentence1": "Over different models, sizes, training methods, and datasets (MNIST and CIFAR-10), we find that (i) in terms of lower bounding the minimum l ∞ adversarial distortion 3 , the optimal layer-wise convex relaxation only slightly improves the lower bound found by , especially when compared with the upper bound provided by the PGD attack, which is consistently 1.5 to 5 times larger; (ii) in terms of upper bounding the robust error, the optimal layer-wise convex relaxation does not significantly close the gap between the PGD lower bound (or MILP exact answer) and the upper bound from .", "sentence2": "there seems to be an inherent barrier blocking our progress on this road of layer-wise convex relaxation, and we hope this work provokes much thought in the community on how to bypass it.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_427", "sentence1": "These establish the conditions, in the context of the variational problem of Section 2, under which each of these algorithms are optimal.", "sentence2": "this allows us to understand the prior assumptions which these algorithms make on the gradients of the objective function they are trying to minimize, and the way noise is introduced in the sampling of stochastic gradients, (gt)t>0.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_428", "sentence1": "The fact that this framework can naturally recover these algorithms begs further study.", "sentence2": "it is still an open question whether it is possible to recover other stochastic algorithms via this framework, particularly those with second-order scaling adjustments such as ADAM or AdaGrad.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_429", "sentence1": "As a result, we can recover algorithms from this framework which have built-in online learning properties.", "sentence2": "these algorithms use an online Bayesian filter on the stream of noisy gradient samples, g t , to compute estimates off (x t ).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_430", "sentence1": "This derivation demonstrates that the solution to the variational problem described in Section 2, under the assumption of a Gaussian model for the evolution of gradients, recovers mirror descent and SGD.", "sentence2": "the martingale gradient model proposed in this section can be roughly interpreted as assuming that gradients behave as random walks over the path of the optimizer.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_431", "sentence1": "The choice of h determines the way we measure distance, and is typically chosen so that it mimics features of the loss function f .", "sentence2": "this quantity plays a central role in mirror descent and non-linear sub-gradient algorithms.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_432", "sentence1": "(2017a), the structure of the manifold interacts with the local approximation of the objective function in a complicated way.", "sentence2": "we provide convergence guarantees for perturbed first order Riemannian optimization methods to seond-order stationary points (local minima).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_433", "sentence1": "It is well-known that convexity/concavity does not imply the PL condition and PL condition does not imply convexity/concavity [30].", "sentence2": "the problems we consider in the next section are neither restriction nor extension of our results on PL games.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_434", "sentence1": "Experimental Setup: The recent work in [42] observed that training a logisitic regression model to classify the images of the Fashion MNIST dataset can be biased against certain categories.", "sentence2": "the function gλ(·)  becomes smooth with Lipschitz gradient; see Lemma B.1 in the supplementary.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_435", "sentence1": "Also note that the notion of FNE does not depend on the ordering of the min and max.", "sentence2": "to be consistent with our notion of PL-games, we can formulate the problem as generative adversarial imitation learning of linear quadratic regulators is an example of finding a FNE for a min-max PL-game.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_436", "sentence1": "In fact, even finding local Nash equilibria is NP-hard in the general non-convex non-concave regime.In addition, as shown by [28, Proposition 10], local Nash equilibria for general non-convex non-concave games may not exist.", "sentence2": "consider the game min Then (0, 0) is the only first-order Nash equilibrium and is not a second-order Nash equilibrium.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_437", "sentence1": "Similarly, one can show that even second-order Nash equilibrium may not exist for non-convex games, see Section 2 for more details.", "sentence2": "a well-justified objective is to find first order Nash equilibria of such games [48]; see definitions and discussion in Section 2.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_438", "sentence1": "Given an `2 N, let m `be the sum of the multiplicities of the first `nonzero top eigenvalues of L K .", "sentence2": "the following corollary follows immediately from Corollary 2.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_439", "sentence1": "This problem is reminiscent of the \"class imbalance\" problem in discriminatory modeling -the challenge however is exacerbated since there is not a single column to balance and the real data distribution should be kept intact.", "sentence2": "the goal is to resample efficiently in a way that all the categories from discrete attributes are sampled evenly (but not necessary uniformly) during the training process, and to recover the (not-resampled) real data distribution during test.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_440", "sentence1": "Since columns in a row do not have local structure, we use fully-connected networks in generator and critic to capture all possible correlations between columns.", "sentence2": "we use two fully-connected hidden layers in both generator and critic.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_441", "sentence1": "Recent works have shown that stochastic gradient descent (SGD) achieves the fast convergence rates of full-batch gradient descent for over-parameterized models satisfying certain interpolation conditions.", "sentence2": "the step-size used in these works depends on unknown quantities and SGD's practical performance heavily relies on the choice of this step-size.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_442", "sentence1": "(ii) The performance of SGD with line-search and Polyak momentum is always better than \"tuned\" constant step-size SGD and Adam, whereas that of SGD with Goldstein line-search is competitive across datasets.", "sentence2": "compared to SGD, checking this condition only makes use of additional mini-batch function (and not gradient) evaluations.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_443", "sentence1": "Note that the above line-search condition uses the function and gradient values of the mini-batch at the current iterate w k .", "sentence2": "compared to SGD, checking this condition only makes use of additional mini-batch function (and not gradient) evaluations.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_444", "sentence1": "The simulation results can be found in Figure 1, and clearly indicate that using greedy planning leads to negligible degradation in the performance.", "sentence2": "the simulations verify our claim that greedy policy updates greatly improve the efficiency of the algorithm while maintaining the same performance.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_445", "sentence1": "In many applications, the high computational complexity of model-based RL makes them infeasible.", "sentence2": "practical model-based approaches alleviate this computational burden by using short-term planning e.g., Dyna [Sutton, 1991], instead of full-planning.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_446", "sentence1": "Note that Theorem 9 exhibits similar problem-dependent regret-bounds as in Theorem 1 of [Zanette and Brunskill, 2019].", "sentence2": "the same corollaries derived in [Zanette and Brunskill, 2019] for EULER can also be applied to EULER-GP.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_447", "sentence1": "Surprisingly, we find that both generalized algorithms do not suffer from performance degradation, up to numerical constants and logarithmic factors.", "sentence2": "we conclude that there exists an RL algorithm that achieves the minimax regret bound, while acting according to greedy policies.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_448", "sentence1": "In this work, we established that tabular model-based RL algorithms can explore by 1-step planning instead of full-planning, without suffering from performance degradation.", "sentence2": "the value V is initialized optimistically and the algorithm interacts with the unknown environment in an episodic manner.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_449", "sentence1": "In this work, we established that tabular model-based RL algorithms can explore by 1-step planning instead of full-planning, without suffering from performance degradation.", "sentence2": "exploring with model-based greedy policies can be minimax optimal in terms of regret.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_450", "sentence1": "Using Lemma 1, we observe that the sequence of values is decreasing and bounded from below.", "sentence2": "intuitively, the decrements of the values cannot be indefinitely large.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_451", "sentence1": "Note that the counterexamples are simply state vectors without labels, and their Lyapunov risk will be determined by the learner, not the falsifier.", "sentence2": "although it is possible to have spurious counterexamples due to the δ error, they are used as extra samples and do not harm correctness of the end result.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_452", "sentence1": "[42] proposes to overcome some of these limitations by combining the UVFA approach with GPI.", "sentence2": "it doesn't formulate a general framework for choosing base policies and their embeddings when learning a particular task space, or for sampling these policies, or addresses the issue of task boundaries and online adjustment of policy use.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_453", "sentence1": "Our algorithm draws inspiration from biology, where animals face similar continual learning tasks while foraging or evading predators in familiar environments.", "sentence2": "we combine the successor representation (SR), that factors the value of actions into expected outcomes and corresponding rewards, with evaluating task similarity through clustering the space of rewards.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_454", "sentence1": "We investigated two approaches that could enable this flexibility: factorized representations, which abstract away general aspects of a task from those prone to change, and nonparametric, memory-based approaches, which can provide a principled way of using similarity to past experiences to guide current behaviour.", "sentence2": "a number of recent or concurrent papers have proposed algorithms for introducing transfer into RL/deep RL settings, by using multiple policies in some way, though none of them use an inferential framework similar to ours, provides a principled way to deal with unsignalled task boundaries, or explains biological phenomena.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_455", "sentence1": "A sharply tuned policy for a previous reward setting with reward locations close to current rewards could lose out to a broadly tuned, but mostly unhelpful competing policy.", "sentence2": "keeping all stored policies diffuse, or otherwise regularising them can be costly, as it can hinder exploitation or the fast instantiation of optimal policies given new rewards.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_456", "sentence1": "The tabular setting enables us to test many components of the algorithm and compare the emerging representations to their biological counterparts.", "sentence2": "it is important to validate that these can be scaled up and used with function approximators, allowing the use of continuous state and action spaces and more complex tasks.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_457", "sentence1": "We investigated two approaches that could enable this flexibility: factorized representations, which abstract away general aspects of a task from those prone to change, and nonparametric, memory-based approaches, which can provide a principled way of using similarity to past experiences to guide current behaviour.", "sentence2": "we combine the successor representation (SR), that factors the value of actions into expected outcomes and corresponding rewards, with evaluating task similarity through clustering the space of rewards.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_458", "sentence1": "In this paper, we show that such independence is in fact not needed for such results which continue to hold under fairly general dependence structures.", "sentence2": "we present uniform bounds on random quadratic forms of stochastic processes which are conditionally independent and sub-Gaussian given another (latent) process.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_459", "sentence1": "In this paper, we consider a generalization of such existing results by allowing for statistical dependence in  ξ.", "sentence2": "we assume  ξ = { ξj } to be a stochastic process where the marginal random variables  ξ j are conditionally independent and sub-Gaussian given some other (latent) process F = {F j }.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_460", "sentence1": "For the uniform bound, the two terms can be separated using Jensen's inequality, the first term can be bounded using Theorem 1 and the second term can be bounded using a standard application of generic chaining using (SP-1) and (SP-2).", "sentence2": "mean shifted versions of our results also hold.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_461", "sentence1": "In the learning process of neural networks, their weight parameters are updated iteratively so that the loss decreases.", "sentence2": "in some settings the loss does not decrease simply, but its decreasing speed slows down significantly partway through learning, and then it speeds up again after a long period of time.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_462", "sentence1": "Deep learning, and neural network as its essential component, has come to be applied to various fields.", "sentence2": "these still remain unclear in various points theoretically.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_463", "sentence1": "This is contrast to the case of linear networks which have no activation; in that case, as µ1 decreases the speed of learning gets exactly 1/µ1 -times larger.", "sentence2": "this phenomenon is peculiar to nonlinear networks.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_464", "sentence1": "D, E, F is the second layers' counterpart of Q, R, T .", "sentence2": "we consider a two-layer perceptron which has N input units, K hidden units and 1 output unit.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_465", "sentence1": "These indicate that the length of the plateau shortens as µ2 becomes large.", "sentence2": "the more the distribution of nonzero eigenvalues gets broaden, the more the plateau gets alleviated.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_466", "sentence1": "We clarified the relationship between the input data statistics and plateau phenomenon.", "sentence2": "it is shown that the data whose covariance matrix has small and disparsed eigenvalues tend to make the phenomenon inconspicuous, by numerically analyzing the macroscopic system.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_467", "sentence1": "We clarified the relationship between the input data statistics and plateau phenomenon.", "sentence2": "the generated data (ξ, t) is then fed to the student network stated above and learned by it in the on-line manner (see Figure 3).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_468", "sentence1": "It indicates that the plateau length and height heavily depend on µ1 , the input scale.", "sentence2": "as µ1 decreases, the plateau rapidly becomes longer and lower.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_469", "sentence1": "We applied our test to check if state-of-the-art classifiers for the ImageNet dataset [8] have been overfitted to the test set.", "sentence2": "we use the VGG16 classifier of [27] and the Resnet50 classifier of [16].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_470", "sentence1": "The generalization bounds are also applicable when the model and the data are dependent (e.g., for cross validation or for error estimates based on the training data or the reused test data), but they usually lead to loose error bounds.", "sentence2": "although much tighter bounds are available if the test data and the model are independent, comparing confidence intervals constructed around the training and test error rates leads to an underpowered test for detecting the dependence of a model on the test set.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_471", "sentence1": "Thus, the test error rate and the adversarial error estimate (calculated based on the same test set) must be close if the test set and the model are independent, and are expected to be different in the opposite case.", "sentence2": "if the gap between the two error estimates is large, the independence hypothesis (i.e., that the model and the test set are independent) is dubious and will be rejected.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_472", "sentence1": "More importantly, since it is based on adversarially generated data points, the adversarial estimator is expected to differ significantly from the test error rate if the model is overfitted to the test set, providing a way to detect test set overfitting.", "sentence2": "the test error rate and the adversarial error estimate (calculated based on the same test set) must be close if the test set and the model are independent, and are expected to be different in the opposite case.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_473", "sentence1": "Under (H), standard concentration inequalities, such as the Chernoff or empirical Bernstein bounds [3], can be used to quantify how fast R S and R g (f ) concentrate around the expected error R(f ).", "sentence2": "we assume that the data is linearly separable with a margin and the density  ρ is known.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_474", "sentence1": "Spatially-varying regularization has also been addressed from a Bayesian view in [34] by putting priors on B-spline transformation parameters.", "sentence2": "metric estimation in [34] is in a fixed atlas-space, whereas RDMM addresses general pairwise image registration.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_475", "sentence1": "In [23] both the velocity field and the regularizer are assumed to be constant in time.", "sentence2": "both are time-dependent in RDMM.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_476", "sentence1": "Theorem 1 (Image-based RDMM optimality conditions).", "sentence2": "regularizer values indicate effective local standard deviations of a local multi-Gaussian regularizer.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_477", "sentence1": "However, static quadratures do not account for the error made during the integration.", "sentence2": "the quadrature is inaccurate when the integrand is not smooth enough and the number of integration steps is too small.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_478", "sentence1": "We observed in our experiments that sharing the same integrand function does not impact performance.", "sentence2": "the neural embedding function h i must produce a fixed size output for i ∈ {1, . . . , d}.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_479", "sentence1": "In this paper, we investigate graphical conditions to allow efficient identification in arbitrary linear structural causal models (SCMs).", "sentence2": "we develop a method to efficiently find unconditioned instrumental subsets, which are generalizations of IVs that can be used to tame the complexity of many canonical algorithms found in the literature.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_480", "sentence1": "Each element of Z has edges to all parents of y, meaning that any scIS existing in the graph must have as many instruments as there are clauses.", "sentence2": "finding an scIS for y in this graph corresponds to finding a satisfying assignment for the formula.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_481", "sentence1": "[38] performs a thorough investigation of various persistent and non-persistent, as well as convergent and non-convergent learning schemes.", "sentence2": "the emphasis is on learning proper energy function with persistent and convergent Markov chains.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_482", "sentence1": "We hypothesize that an ideal unsupervised learning algorithm should use past observations to create a stable representation of the environment.", "sentence2": "a representation that captures the global factors of variation of the environment in a temporally coherent way.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_483", "sentence1": "The challenge of model-based RL in rich 3D environments has compelled some researchers to use privileged information such as camera-locations [15], depth information [53], and other ground-truth state variables of the state simulator [54,49].", "sentence2": "some work has provided evidence that we may not need very expressive models to benefit to some degree [30].", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_484", "sentence1": "This state is a belief state and is used to calculate policy and value as well as being the starting point for predictions of the future.", "sentence2": "a representation that captures the global factors of variation of the environment in a temporally coherent way.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_485", "sentence1": "For both datasets, as the number of layers (hence number of parameters) increases, overfitting becomes more severe for no tying.", "sentence2": "in the above regression tasks, the performance metric is the MLL of the test data (or test MLL).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_486", "sentence1": "Additional results for IPVI with and without parameter tying are found in Appendix C.3.", "sentence2": "parameter tying alleviates the overfitting considerably.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_487", "sentence1": "For both datasets, as the number of layers (hence number of parameters) increases, overfitting becomes more severe for no tying.", "sentence2": "parameter tying alleviates the overfitting considerably.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_488", "sentence1": "Since then, several extensions to the general methodology have been proposed (Ahmed et al., 2017;Shen et al., 2018).", "sentence2": "it is still not entirely clear: what do the multiple heads in these models buy us?", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_489", "sentence1": "As a first step to understand whether some heads are universally important, we perform the same ablation study on a second, out-of-domain test set.", "sentence2": "we consider the MNLI \"mismatched\" validation set for BERT and the MTNT English to French test set (Michel and Neubig, 2018) for the WMT model, both of which have been assembled for the very purpose of providing contrastive, out-of-domain test suites for their respective tasks.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_490", "sentence1": "Attention is a powerful and ubiquitous mechanism for allowing neural models to focus on particular salient pieces of information by taking their weighted average when making predictions.", "sentence2": "multi-headed attention is a driving force behind many recent state-of-the-art natural language processing (NLP) models such as Transformer-based MT models and BERT.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_491", "sentence1": "This observation begets the question: is more than one head even needed?", "sentence2": "we compute the difference in performance when all heads except one are removed, within a single layer.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_492", "sentence1": "Figure 4 shows that performance drops much more rapidly when heads are pruned from the Enc-Dec attention layers.", "sentence2": "pruning more than 60% of the Enc-Dec attention heads will result in catastrophic performance degradation, while the encoder and decoder self-attention layers can still produce reasonable translations (with BLEU scores around 30) with only 20% of the original attention heads.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_493", "sentence1": "In particular, pruning more than 60% of the Enc-Dec attention heads will result in catastrophic performance degradation, while the encoder and decoder self-attention layers can still produce reasonable translations (with BLEU scores around 30) with only 20% of the original attention heads.", "sentence2": "encoder-decoder attention is much more dependent on multi-headedness than self-attention.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_494", "sentence1": "Since then, several extensions to the general methodology have been proposed (Ahmed et al., 2017;Shen et al., 2018).", "sentence2": "this suggests that the important heads are determined early (but not immediately) during the training process.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_495", "sentence1": "Unfortunately, in our case, this projection is quite complicated, since the intersection of the sparsity constraint and the quantization constraint is complicated.", "sentence2": "we notice that the projection onto the feasible set defined by each individual constraint is doable (the projection onto the sparsity constraint is quite standard, how to do efficient projection onto the quantization constraint defined set will be clear soon).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_496", "sentence1": "The enormous complexity of CNNs remains a major inhibitor for their more extensive applications in resource-constrained IoT systems.", "sentence2": "model compression [5] is becoming increasingly demanded and studied [6][7][8][9].", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_497", "sentence1": "Meanwhile, ATMC also achieves very close, sometimes better accuracy-compression ratio trade-offs on benign testing sets than NAP, with much enhanced robustness.", "sentence2": "it has indeed combined the best of both worlds.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_498", "sentence1": "We next briefly review three mainstream compression methods: pruning, factorization, and quantization.", "sentence2": "we recognize that using group sparsity norms in (3) might potentially be a preferred option if ATMC will be adapted for model acceleration.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_499", "sentence1": "Next, since the target model needs to defend against the attacker, it requires to suppress the worst risk.", "sentence2": "the overall objective for the target model to gain adversarial robustness could be expressed as Z denotes the training data set: As we reviewed previously, typical CNN model compression strategies include pruning (element-level [10], or channel-level [8]), low-rank factorization [16,13,17], and quantization [23,19,20].", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_500", "sentence1": "To demonstrate that ATMC achieves remarkably favorable trade-offs between robustness and model compactness, we carefully design experiments on a variety of popular datasets and models as summarized in Section 3.1.", "sentence2": "since no algorithm with exactly the same goal (adversarially robust compression) exists off-the-shelf, we craft various ablation baselines, by sequentially composing different compression strategies with the state-of-the-art adversarial training [32].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_501", "sentence1": "On all three victim models, our method significantly outperforms NES and Bandits-TD in both query efficiency and success rates.", "sentence2": "prior Gradients as Basis Vectors?", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_502", "sentence1": "As introduced in the previous section, we reckon that it is promising to apply the gradient of some reference models to span the search subspace for mounting black-box attacks.", "sentence2": "there remain some challenges in doing so.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_503", "sentence1": "As suggested in Figure 2(b), one way of guaranteeing a low failure rate in our method is to collect adequate reference models.", "sentence2": "we know from prior works that even adversarial examples crafted using some single-step attacks like the fast gradient (sign) [18] can transfer [28,22], hence one can hypothesize that the gradients of some \"substitute\" models are more helpful in spanning the search subspaces with reduced dimensionalities.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_504", "sentence1": "As suggested in Figure 2(b), one way of guaranteeing a low failure rate in our method is to collect adequate reference models.", "sentence2": "it is usually troublesome in practice, if not infeasible.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_505", "sentence1": "Our solution to resolve this issue is inspired by the dropout [35] and \"droplayer\" (a.k.a., stochastic depth) [12] techniques.", "sentence2": "it is usually troublesome in practice, if not infeasible.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_506", "sentence1": "Additionally, the proposed transformation of the covariance function seems to be better suited for ordinal-valued variables rather than categorical variables, further restricting the utility of this approach.", "sentence2": "we propose a method that can deal with high-dimensional combinatorial (categorical and/or ordinal) spaces.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_507", "sentence1": "If two vertices are connected by an edge, then their respective set of combinatorial choices differ only by a single combinatorial choice.", "sentence2": "we can now revisit the notion of smoothness on combinatorial structures as smoothness of a graph signal [8,35] defined on the combinatorial graph.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_508", "sentence1": "We define then smoothness of functions on combinatorial structures to be the smoothness of graph signals using the Graph Fourier Transform (GFT) [35].", "sentence2": "we propose as our GP kernel on the graph a variant of the diffusion kernel, the automatic relevance determination(ARD) diffusion kernel, for which computing the GFT is computationally tractable via a decomposition of the eigensystem.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_509", "sentence1": "We circumvent this requirement by the notion of the combinatorial graph defined as a graph, which contains all possible combinatorial choices as its vertices for a given combinatorial problem.", "sentence2": "each vertex corresponds to a different joint assignment of categorical or ordinal variables.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_510", "sentence1": "When comparing min(ESS)/s, SA outperforms AM by 31x on covertype and 11x on MiniBoonE and outperforms NUTS by 24x on covertype and 147x on MiniBoonE.", "sentence2": "sA samples effectively from this high-dimensional, challenging posterior distribution without requiring any tuning of the initialization distribution.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_511", "sentence1": "There are a few work related to our considered class of OT with tree metrics [35,62].", "sentence2": "kloeckner [35] studied geometric properties of OT space for measures on an ultrametric space, and Sommerfeld and Munk [62] focused on statistical inference for empirical OT on finite spaces including tree metrics.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_512", "sentence1": "We use the farthest-point clustering due to its fast computation.", "sentence2": "the complexity of the farthest-point clustering into κ clusters for n data points is O(n log κ) using the algorithm in [23]. ", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_513", "sentence1": "As in [42], we generated 1000 orbits for each class where each orbit contains 1000 points.", "sentence2": "we propose a positive definite tree-(sliced-)Wasserstein kernel that generalizes the sliced-Wasserstein kernel [11,36].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_514", "sentence1": "Following [33, §1, p.245-247], an ultrametric implies a tree structure which can be constructed by hierarchical clustering schemes.", "sentence2": "an ultrametric is a tree metric.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_515", "sentence1": "The TW distance, as well as its average over several trees, can be shown to be negative definite 1 .", "sentence2": "we propose a positive definite tree-(sliced-)Wasserstein kernel that generalizes the sliced-Wasserstein kernel [11,36].", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_516", "sentence1": "Louizos et al. [2017] studied scale mixtures of Gaussian priors and half-Cauchy scale priors for the hidden units of VGG models [Simonyan and Zisserman, 2014] and achieved good model compression performance on CIFAR10 [Krizhevsky, 2009] using VI.", "sentence2": "due to the limitation of VI in non-convex optimization, the compression is still not sparse enough and can be further optimized.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_517", "sentence1": "Ghosh et al. [2018] proposed to use variational inference (VI) based on regularized horseshoe priors to obtain a compact model.", "sentence2": "all the analyses fail when DNN has too many parameters, and the over-specified model tends to have a large prediction variance, resulting in poor generalization and causing over-fitting.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_518", "sentence1": "However, all the analyses fail when DNN has too many parameters, and the over-specified model tends to have a large prediction variance, resulting in poor generalization and causing over-fitting.", "sentence2": "a proper model selection is on demand at this situation.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_519", "sentence1": "In all of these applications, because a sample is needed at every epoch t, it is desirable to have a fast online sampling algorithm.", "sentence2": "the ultimate goal is to design an algorithm for Problem 1.1 such that the number of gradient evaluations is almost constant at each epoch t, so that the computational requirements at each epoch do not increase over time.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_520", "sentence1": "To overcome this, we could modify our algorithm by including an adaptive pre-conditioner which changes along with the target distribution.", "sentence2": "there are many streaming algorithms that are used in practice which lack provable guarantees, or which rely on properties of the data (such as compressibility [HCB16,CB19]).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_521", "sentence1": "Motivated by real-world applications, we assume that functions are smooth, their associated distributions have a bounded second moment, and their minimizer drifts in a bounded manner, but do not assume they are strongly convex.", "sentence2": "our assumptions hold for online Bayesian logistic regression, when the data satisfy natural regularity properties, giving a sampling algorithm with updates that are poly-logarithmic in T .", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_522", "sentence1": "These problems lead us to a relaxed implementation of (Obj 0 ) , bringing us possibilities to embrace a practical and a more flexible solution.", "sentence2": "the solution produced by the optimization method shows a close connection between our method and the optimal transport problem, which brings new insight into how negative transfer could be prevented across latent tasks and output tasks.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_523", "sentence1": "From a non-asymptotic perspective, Prop.2-(b) shows how fast this approximation will take place.", "sentence2": "we will get a reasonable approximation of the original OT problem with a small ϑ.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_524", "sentence1": "The N-gram approach constructs a representation vector c (n) for the sentence, whose coordinates correspond to all n-grams and the value of a coordinate is the number of times the corresponding n-gram shows up in the sentence.", "sentence2": "the dimension of an n-gram vector is |V | n for a vocabulary V , and the vector c (1) is just the count vector of the words in the sentence.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_525", "sentence1": "In contrast, virtual screening (i.e., selecting molecules based on predicted properties via machine learning methods) can be done in minutes for predicting millions of molecules.", "sentence2": "it can be a good filtering step before the physical experiments, to help accelerate the drug discovery process and significantly reduce resource requirements.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_526", "sentence1": "At iteration n, each vertex updates its latent vector by element-wise multiplying it with the sum of the latent vectors of its neighbors.", "sentence2": "at the end of iteration n, the latent vector on vertex i is the sum of the embeddings of the walks that end at i and have length n, and the sum of the all latent vectors is the embedding of the n-gram walk set (with proper scaling).", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_527", "sentence1": "Since our focus is to compare the representations of the graphs, no transfer learning or multi-task learning is considered.", "sentence2": "we are comparing each task independently, which gives us 28 regression tasks and 32 classification tasks in total.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_528", "sentence1": "Deep Tensor Neural Networks (DTNN) [47] and Message-Passing Neural Networks (MPNN) [26] are two graph neural networks that are able to utilize 3D information encoded in the datasets.", "sentence2": "we further compare our method to these two most advanced GNN models, on the two datasets QM8 and QM9 that have 3D information.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_529", "sentence1": "Finally as has been done in previous works, e.g. .g. [5, 8–10] for compressed sensing and denoising, translating our results to practical situations in designing an AMP algorithm that takes care of correlated GAN or VAE weights is still under investigatio", "sentence2": "we solve the fixed point equations (12) and plot the MMSE obtained from the fixed point in a heat map, for the linear, sign and relu activations.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_530", "sentence1": "We analyze the Bayesoptimal performance under specific generative models for the spike.", "sentence2": "as a unique fixed point is found, the Bayes optimal errors are continuous and we did not observe any algorithmic gap.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_531", "sentence1": "These results demonstrate that the exponential scheme convicingly outperforms the polynomial step-size schemes.", "sentence2": "we consider the use of: Where, we perform a systematic grid search on the parameters η 0 and b.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_532", "sentence1": "In contrast to the non-smooth setting, the state of our understanding of SGD's final iterate for smooth stochastic convex optimization, or, say, the streaming least squares regression setting is far less mature − this gap motivates our paper's contributions.", "sentence2": "practical SGD implementations typically return the final iterate of a stochastic gradient procedure.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_533", "sentence1": "In contrast, this paper shows that Step Decay schedules, which cut the learning rate by a constant factor every constant number of epochs (i.e., the learning rate decays geometrically) offer significant improvements over any polynomially decaying step size schedule.", "sentence2": "the behavior of the final iterate with step decay schedules is off from the statistical minimax rate by only log factors (in the condition number for the strongly convex case, and in T in the non-strongly convex case).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_534", "sentence1": "First introduced by Ioffe and Szegedy [2015], BatchNorm fixes layer distributions to reduce ICS (Internal Covariate Shift), a phenomenon that the upper layers need to continuously adapt to the new distributions of lower layers.", "sentence2": "eq. ( 3) shows that normalization re-centers and re-scales input vector x.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_535", "sentence1": "The improvement mainly comes from the stronger expressive power of deep layers.", "sentence2": "with the increase of depth, the network training process becomes complicated and requires advanced architectural techniques.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_536", "sentence1": "Initially, considering that input information may be lost when normalizing input distributions, the bias and gain are designed for affine transformation on normalized vectors to enhance the expressive power.", "sentence2": "since the bias and gain are learned from the training set and they ignore the input distributions of the testing data, the risk of over-fitting may increase in LayerNorm.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_537", "sentence1": "The improvement mainly comes from the stronger expressive power of deep layers.", "sentence2": "a typical example is its application in the state-of-the-art framework, Transformer [Vaswani et al., 2017].", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_538", "sentence1": "Next we focus on the theoretical lower bound for the efficiency-fairness tradeoff that any algorithm could achieve.", "sentence2": "we show that the tradeoff achieved in Theorem 3.1 is actually tight in this model.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_539", "sentence1": "Leaving the system as it is would affect the sustainability of the ridesourcing business model in the long run, as unsatisfied drivers will not renew their memberships and new drivers will be deterred from signing up.", "sentence2": "fairness on the drivers' side should be assessed more carefully and should receive more attention.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_540", "sentence1": "Efficiency and fairness are often competing objectives such that in most cases, the optimum of both cannot be achieved simultaneously.", "sentence2": "we are naturally led to the question of how to reconcile system efficiency and drivers' fairness in a ridesourcing assignment.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_541", "sentence1": "In this paper, we focus on both the system efficiency and the fairness among drivers and quantitatively analyze the tradeoffs between these two objectives.", "sentence2": "we give an explicit answer to the question of whether there always exists an assignment that achieves any target efficiency and fairness.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_542", "sentence1": "Note that such changes are two-way (i.e., efficiency or fairness-oriented), gradual, and dependent on dynamically changing demand conditions.", "sentence2": "to provide better managerial flexibilities to decision makers, we need to provide a full set of candidate allocation solutions and characterize the trade-offs inherent in applying these concepts.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_543", "sentence1": "For instance, a central decision maker may choose a mild strategy that balances the tension between efficiency and fairness in normal traffic conditions.", "sentence2": "as demand increases and traffic conditions worsen, it may be desirable to move to a strategy that puts higher attention to efficiency to quickly serve the waiting passengers.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_544", "sentence1": "To reflect real road conditions and traveling time, we construct a road network of Manhattan with 3,671 nodes and 7,674 edges.", "sentence2": "it is evident that our algorithm manages to redistribute the trip utility increments to the vehicles with low historical utilities, without sacrificing too much on the efficiency.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_545", "sentence1": "[5,6] has specifically studied fairness and the corresponding efficiency loss in a general divisible resource allocation framework and applied their results to a case study in the context of air traffic management.", "sentence2": "none of these works can easily incorporate additional waiting time and pick-up distance constraints because these are unique to the ridesourcing problem.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_546", "sentence1": "(i), (ii), and (iii) are intricately connected and may have competing effect up to some extent on the efficiency loss.", "sentence2": "we can conclude that the theoretical worst-case efficiency loss does not necessarily arise even in artificial examples; here, we have seen the problem instances that show much more benign behaviour.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_547", "sentence1": "While we focused on the correspondence problem between transcriptomics and electrophysiology(k = 2), we presented the technical development of k-coupled autoencoders in full generality.", "sentence2": "our method is applicable to the joint alignment of additional modalities.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_548", "sentence1": "2B suggests that the individual cell types are well approximated by hyperellipsoids.", "sentence2": "fitting a Gaussian mixture model to the encodings provides an efficient prior distribution for p(zse) (or p(zse)), and produces a generative model for multi-modal datasets.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_549", "sentence1": "The trained network is expected to align the representations without any explicit transformation matrices.", "sentence2": "in the absence of any normalization of the representations, the individual agents can arbitrarily scale down their representations to minimize the coupling cost without a penalty on reconstruction accuracy.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_550", "sentence1": "Parameters of the resulting autoencoding architectures are fitted with stochastic mini-batch training and the Adam optimizer [24].", "sentence2": "in the absence of any normalization of the representations, the individual agents can arbitrarily scale down their representations to minimize the coupling cost without a penalty on reconstruction accuracy.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_551", "sentence1": "We present a rigorous analysis of these problems, and show that normalization with the full covariance matrix of the mini-batch is sufficient, as expected [8], to obtain reasonable latent space representations.", "sentence2": "this calculation can be prohibitively inaccurate depending on the latent space dimensionality and batch size (\"curse of dimensionality\").", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_552", "sentence1": "However, this calculation can be prohibitively inaccurate depending on the latent space dimensionality and batch size (\"curse of dimensionality\").", "sentence2": "we propose an alternative normalization that relies only on estimating the minimum eigenvalue of this covariance matrix.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_553", "sentence1": "In this study, we have demonstrated a surprising level of cross-modal predictive ability across transcriptomic and electrophysiological recordings.", "sentence2": "we showed that the transcriptomic class can be predicted with ∼80% accuracy from electrophysiological recordings when the transcriptomic hierarchy is resolved into 15 classes, and with ∼70% accuracy when it is resolved into 25 classes (λ = 10 results).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_554", "sentence1": "Xu et al. [36] and Gleave and Habryka [12] have proposed approaches that combine IRL and gradient-based meta-learning [9], which provide promising results on deriving generalizable reward functions.", "sentence2": "they have been limited to tabular MDPs [36] or settings with provided task distributions [12], which are challenging to gather in real-world applications.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_555", "sentence1": "Without further constraints over m, directly applying AIRL to learning the reward function (by augmenting each component of AIRL with an additional context variable m inferred by q ψ ) could simply ignore m, which is similar to the case of InfoGAIL [21].", "sentence2": "some connection between the reward function and the latent context variable m need to be established.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_556", "sentence1": "The primary contribution of this paper is a new framework, termed Probabilistic Embeddings for Meta-Inverse Reinforcement Learning (PEMIRL), which enables meta-learning of rewards from unstructured multi-task demonstrations.", "sentence2": "pEMIRL combines and integrates ideas from context-based meta-learning [5,26], deep latent variable generative models [17], and maximum entropy inverse RL [42,41], into a unified graphical model (see Figure 4 in Appendix D) that bridges the gap between few-shot reward inference and learning from unstructured, heterogeneous demonstrations.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_557", "sentence1": "The coherence curves follow similar trend for other values for the threshold 2 ≤ T ≤ 20.", "sentence2": "the measure is sensitive to topic sparsity; if the support of the topic distribution is smaller than T the measure becomes more noisy and less meaningful.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_558", "sentence1": "The topics for our model are semantically meaningful, capturing certain intuitive themes, as desired.", "sentence2": "the model applies KLdivergences between the observed similarities (distributions) and latent distributions as likelihoods.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_559", "sentence1": "The saturating distance curves also show an effective range for λ values; for λ >> 0.2 (not shown), the computations eventually become more unstable, because more and more terms are assigned to the empirical distributions and the topics become too sparse complicating posterior inference.", "sentence2": "the entropies show how the topics become (on average) more sparse for increasing λ.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_560", "sentence1": "The K-divergence is always well-defined for all values for qm ∈ ∆; this is especially relevant at the boundaries of ∆.", "sentence2": "the K-divergence (4) is not as sensitive to misses as KL(p m , q m ), which approaches infinity close to the boundaries imposing infinite penalty for misses, essentially, imposing a barrier function.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_561", "sentence1": "Indexing starts at 1, with A i,: and A :,i indicating the i-th row and i-th column of A respectively.", "sentence2": "now the entries of vector w + are in between 0 and 1.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_562", "sentence1": "That is, if M = L in Theorem 1 then it becomes the known Cheeger's inequality.", "sentence2": "lemma 2 (Matrix Bernstein inequality, Theorem 1.4 in (Tropp 2012)).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_563", "sentence1": "In contrast to (Globerson et al. 2015, Foster et al. 2018), we study the sufficient conditions for exact recovery in polynomial time, and provide high probability results for general families of undirected connected graphs, which we consider to be a novel result to the best of our knowledge.", "sentence2": "we show that weak-expander graphs (e.g., grids) can be exactly recovered by adding small perturbations (edges coming from the Erdos-Rényi model with small probability).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_564", "sentence1": "Since our prediction is a column vector y, we have that y y is rank-1 and symmetric, which implies that Y is a positive semidefinite matrix.", "sentence2": "our relaxation to the combinatorial problem (3) results in the following primal formulation 2 : We will make use of the following matrix concentration inequality for our main proof.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_565", "sentence1": "This fact combined with complementary slackness, primal and dual feasibility, entail that Y is a multiple of y y .", "sentence2": "we must have that Y = y y because Y i,i = 1.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_566", "sentence1": "Since we have that y is an eigenvector of V − X with eigenvalue zero, showing that λ 2 (V − X) > 0 will imply that V − X is positive semidefinite.", "sentence2": "we focus on controlling its second smallest eigenvalue.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_567", "sentence1": "That is, if M = L in Theorem 1 then it becomes the known Cheeger's inequality.", "sentence2": "our result in Theorem 1 apply for more general matrices and is of use for our next result.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_568", "sentence1": "Our results show that exact recovery is possible and achievable in polynomial time for a large class of graphs.", "sentence2": "we show that graphs that are bad expanders can be exactly recovered by adding small edge perturbations coming from the Erdős-Rényi model.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_569", "sentence1": "Finally, from Theorem 3, it is clear that since the parameter q ∈ (0, 0., for a sufficiently large n we have an exponential decay of the probability of error 2 .", "sentence2": "we focus on the conditions of the first stage and provide examples in the next section.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_570", "sentence1": "When keeping more keypoints, poor local maxima starts to get selected for these models (e.g in the sky or the river in Figure 4) and the matching performance drops.", "sentence2": "having numerous keypoints is important for many applications such as visual localization because it augments the chance that at least a few of them will be correctly matched despite occlusions or other noise sources.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_571", "sentence1": "After learning the keypoint detector, a deep descriptor is trained using a second network branch, sharing most of the computation.", "sentence2": "our approach learns both of them jointly from scratch and without introducing any artificial bias in the keypoint detector, which is also achieved by Georgakis et al.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_572", "sentence1": "We note that the Baseline model does not locate any keypoints in the arms area.", "sentence2": "when the pose difference with the initial pose increases, the model cannot reconstruct the video (columns 3,4 and 5).", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_573", "sentence1": "Once trained on a set of videos depicting objects of the same category (e.g. faces, human bodies), our method can be applied to any object of this class.", "sentence2": "we note that, with respect to the Baseline model, the MKR of the full model is smaller by the factor of 2.75.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_574", "sentence1": "At test time we apply our model to pairs composed of the source image and of each frame of the driving video and perform image animation of the source object.", "sentence2": "we use equal loss weights in all our experiments.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_575", "sentence1": "Recently, deep generative models have emerged as effective techniques for image animation and video retargeting [2,41,3,42,27,28,37,40,31,21].", "sentence2": "generative Adversarial Networks (GANs) [14] and Variational Auto-Encoders (VAEs) [20] have been used to transfer facial expressions [37] or motion patterns [3] between human subjects in videos.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_576", "sentence1": "In the case of Tai-Chi-HD, the human-pose estimator returns an additional binary label for each keypoint indicating whether or not the keypoints were successfully detected.", "sentence2": "we also report the MKR defined as the percentage of keypoints that are detected in the ground truth frame but not in the generated one.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_577", "sentence1": "Finally, in our preliminary experiments, we observed that our model shows low sensitivity to the relative weights of the reconstruction and the two equivariance losses.", "sentence2": "we use equal loss weights in all our experiments.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_578", "sentence1": "In generating adversarial examples, the gradient is usually normalized [13,24], such that the direction of the gradient estimator, instead of the magnitude, will affect the performance of attacks.", "sentence2": "we also estimate the gradient norm every 10 attack iterations in all experiments to reduce the number of queries, since usually its value is relatively stable in the optimization process.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_579", "sentence1": "These methods all use queries to obtain knowledge of the black-box model, and train/find surrogate models to generate adversarial examples, with the purpose of improving the transferability.", "sentence2": "we do not optimize the surrogate model, but focus on utilizing the gradient of a fixed surrogate model to obtain a more accurate gradient estimate.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_580", "sentence1": "In generating adversarial examples, the gradient is usually normalized [13,24], such that the direction of the gradient estimator, instead of the magnitude, will affect the performance of attacks.", "sentence2": "we incorporate a scaling factor b in Eq. (7) and minimize the error w.r.t. b, which can neglect the impact of the magnitude on the loss of the estimator gˆ.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_581", "sentence1": "In [25], communities are used to re-identify multiple addresses belonging to a same user in Bitcoin trading networks.", "sentence2": "there is a need to hide the community affiliations in order to preserve the privacy of online users.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_582", "sentence1": "Discovering hidden patterns in this network structure is a compelling application of graph data mining algorithms.", "sentence2": "community detection stands out as one of the most important graph mining methods [11,16,23,26].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_583", "sentence1": "Our experimental results obtained on three datasets show that this principle works.", "sentence2": "our segmentation model is competitive with supervised approaches trained on a few hundred labeled examples.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_584", "sentence1": "State-of-the-art methods use clever architectural choices or pipelines tailored to the challenges of the task [5,20,58].", "sentence2": "most of those models use pixel-level supervision, which can be unavailable in some settings, or time-consuming to acquire in any case.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_585", "sentence1": "Note that our objective is not to generate appealing images but to learn an object segmentation function.", "sentence2": "reDO generates images that are less realistic than the ones generated by state-of-the-art GANs.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_586", "sentence1": "If A t (defined in Section 4) is small then it is likely that A t+1 is also small.", "sentence2": "our algorithm is prone to select the last part of iterates.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_587", "sentence1": "A basic and representative method of this type is the gradient descent, which iteratively moves iterates along the minus gradient direction of the current iterate.", "sentence2": "gradient descent applied to machine learning problems requires to go through all training examples at each iteration, which is not efficient when the sample size is large.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_588", "sentence1": "Indeed, one can randomly draw the output according to a distribution over the iterate sequence with the probability mass function determined by any optimal weighted averaging scheme [20,31,37].", "sentence2": "this scheme introduces new variances due to the choice of a random iterate as the output.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_589", "sentence1": "Intuitively, A t is related to the distance between w t and w t+1 .", "sentence2": "our selection of the iterate shares some spirit with the widely used heuristic of terminating the algorithm when the successive iterates are close.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_590", "sentence1": "The choice of algorithms would affect the implicit regularization introduced in the learned models.", "sentence2": "the optimization technique itself \"biases\" towards a certain model in an implicit way (Soudry et al.  [2018]).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_591", "sentence1": "Moreover, [28,27] take a wide spectrum by considering both the social preference and individual variations simultaneously.", "sentence2": "it designs a basic linear mixed-effect model which not only can derive the common preference on population-level, but also can estimate user's preference/utility deviation in an individual-level.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_592", "sentence1": "If a voter thinks college V 3 is better than college V 6 , a solid arrowed line from V 3 to V 6 occurs (i.e., superiority).", "sentence2": "when a voter thinks the two colleges (i.e., V 1 and V 3 ) listed are incomparable and difficult to judge, he may click the button \"I can't decide\", then a dotted line connecting V 1 and V 3 happens (i.e., tie).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_593", "sentence1": "In this paper, instead of learning a global ranking which is agreed with the consensus, we pursue the tie-aware partial ranking from an individualized perspective.", "sentence2": "we formulate a unified framework which not only can be used for individualized partial ranking prediction, but also be helpful for abnormal user selection.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_594", "sentence1": "This is realized by a variable splittingbased algorithm called iSplitLBI.", "sentence2": "our algorithm generates a sequence of estimations with a regularization path, where both the hyperparameters and model parameters are updated.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_595", "sentence1": "Since α-cut is an ensemble-based algorithm, its performance depends on the choice of weak learners.", "sentence2": "we compare our proposed algorithm with the α-cut algorithm where different types of such weak learners and regularization schemes are adopted.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_596", "sentence1": "Clearly one can see that users jumped out earlier (i.e., the top 10% marked with pink) show larger L 2 -distance, thus are those with large deviation from the population's opinion and can be treated as abnormal users.", "sentence2": "users jumped out later (i.e., the bottom 10% marked with blue) tend to have smaller or even zero L 2 -distance.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_597", "sentence1": "Our approach ideally requires identifying important regions that a human considers most critical in answering the question.", "sentence2": "the base system utilizes a Faster R-CNN head [22] in conjunction with a ResNet-101 base network [10] as the object detection module.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_598", "sentence1": "In this section, we quantitatively evaluate the effectiveness of the proposed self-critical objective.", "sentence2": "we evaluate the fraction of false sensitivity where the predicted incorrect answer's sensitivity to the influential object (to which the correct answer is most sensitive) is greater than the Figure 4: Positive examples are showing that our self-critical reasoning approach prevents the incorrectly predicted answer in the UpDn baseline system from being sensitive to the most influential object.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_599", "sentence1": "The state-of-the-art VQA systems [8,2,1,3,12,33,25,29,14,15,21] achieve high performance when the training and test question-answer (QA) pairs are sampled from the same distribution.", "sentence2": "most of these systems fail to generalize to test data with a substantially different QA distribution.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_600", "sentence1": "For question embedding, following [2], we perform standard text pre-processing and tokenization.", "sentence2": "questions are first converted to lower case and then trimmed to a maximum of 14 words, and the words that appear less than 5 times are replaced with an \"<unk>\" token.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_601", "sentence1": "This is because negative gradients on the inputs to a ReLU are valuable evidence against the current prediction.", "sentence2": "there is no need to zero them out with a ReLU.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_602", "sentence1": "In order to ensure the training and test data have different category distributions, we intentionally assign different weights to the two components.", "sentence2": "during training, the examples are drawn from N 1 with probability p, and during test, the examples are drawn from N 1 with probability 1 − p. We examine the effectiveness of our self-critical approach varying p from 0.05 to 0.5 (i.e. 0.05, 0.1, 0.2, 0.5) (0.5 means no train/test difference).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_603", "sentence1": "The number around the bounding box is the answer's sensitivity to the object.", "sentence2": "the work most related to ours is HINT [25], which enforces the system's gradient-based importance scores for each detected object to have the same rankings as its human importance scores.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_604", "sentence1": "First, like the previous work [28,35,25,20], we used regions that humans have explicitly marked as important.", "sentence2": "this requires a fair bit of extra human effort to provide such detailed annotations.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_605", "sentence1": "For example, in VQA-CP, bedrooms are the most common room type.", "sentence2": "during testing, systems frequently incorrectly classify bathrooms (which are rare in the training data) as bedrooms.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_606", "sentence1": "These systems are trained to encourage the network to focus on regions in the image that humans have somehow annotated as important (which we will refer to as \"important regions.\").", "sentence2": "many times, the network also focuses on these important regions even when it produces a wrong answer.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_607", "sentence1": "We showed that by correctly incorporating an estimate of the propensity score into the covariance kernel, one can substantially improve the precision of both the posterior mean and posterior uncertainty quantification.", "sentence2": "this makes the modified GP method highly competitive with state-of-the-art methods.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_608", "sentence1": "Acknowledgements: Botond Szabó received funding from the Netherlands Organization for Scientific Research (NWO) under Project number: 639.031.654.", "sentence2": "more concretely, one can sample from the marginal posterior for ψ by drawing a full posterior sample (F, π, m) and computing the corresponding draw ψ according to the formula (2).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_609", "sentence1": "Then, invariant descriptors can be extracted on the detected local patches.", "sentence2": "finally, GAS uses the average of 128-dimensional embedded group features weighted by the attention weights as descriptors.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_610", "sentence1": "The robustness of detectors can be guaranteed theoretically, e.g., by the scale-space theory [34].", "sentence2": "a typical image often have very few pixels for which viewpoint covariant patches may be reliably detected [22].", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_611", "sentence1": "The Universal Correspondence Network (UCN) [7] uses a convolutional spatial transformer [26] in the network to normalize the local patches to a canonical shape.", "sentence2": "learning an invariant spatial transformer is as difficult as learning a viewpoint covariant detector.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_612", "sentence1": "Such a permutation preserves local structures of the group features.", "sentence2": "we propose to use group convolutions to encode the local structures of the group features, resulting in feature representations that are not only discriminative but equivariant to the transformations in the group.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_613", "sentence1": "We further demonstrate the robustness of GIFT to extremely large scale and orientation changes on several new datasets.", "sentence2": "this transformation-invariant dense descriptor simplifies correspondence estimation as detecting covariant patches can be avoided.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_614", "sentence1": "Some recent works [45,61,29,41,64,12,62] try to learn such viewpoint covariant patch detectors by CNNs.", "sentence2": "the definition of a canonical scale or orientation is ambiguous.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_615", "sentence1": "Instead of relying on a sparse set of covariant patches, some recent works [20,7,11] propose to extract dense descriptors by feeding the whole image into a convolutional neural network (CNN) and constructing pixel-wise descriptors from the feature maps of the CNN.", "sentence2": "the CNN-based descriptors are usually sensitive to viewpoint changes as convolutions are inherently not invariant to geometric transformations.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_616", "sentence1": "The main idea is to add adversarial examples into the training set to improve the robustness.", "sentence2": "earlier work usually only adds adversarial example once or only few times during the training phase.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_617", "sentence1": "Recall that L A (W) is the loss suffered with respect to the perturbation function A.", "sentence2": "this makes f (W, x) = 0 at initialization, which helps eliminate some unnecessary technical nuisance.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_618", "sentence1": "An adversarial example can be generated by maximizing the loss function within an -ball around a natural sample.", "sentence2": "generating adversarial examples can be viewed as solving a constrained optimization problem and can be (approximately) solved by a projected gradient descent (PGD) method [31].", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_619", "sentence1": "Our proof idea utilizes the same high-level intuition as [1,27,18,55,10,11] that near the initialization the network is linear.", "sentence2": "compared with Equation ( 7) which shows an n-interpolation class can be realized by a network architecture with VC-Dimension O(n log n), we can conclude that robust interpolation by neural networks needs more capacity, so increasing the width of neural network is indeed in some sense necessary.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_620", "sentence1": "As an example, using quadratic ReLU as activation function, we solve the explicit dependency on in Appendix C.2 that doesn't rely on Assumption 5.2.", "sentence2": "adversarial training is guaranteed to find a robust classifier under a given attack algorithm when the network width is sufficiently large.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_621", "sentence1": "In contrast, there are neural net architectures that can interpolate n samples with only O(n) parameters and VC-Dimension at most O(n log n).", "sentence2": "the capacity required for robust learning is higher.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_622", "sentence1": "The second approach is to instead learn a representation that is faithful to the underlying data structure, hoping that this is sufficient to disentangle the representation.", "sentence2": "there is currently little to no agreement in the literature on how to learn such representations [Locatello et al., 2019].", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_623", "sentence1": "Inspired by Detlefsen et al. [2018], we also consider transformations T using the highly expressive diffiomorphic transformations CPAB from Freifeld et al. [2015].", "sentence2": "the naming comes from Eq. 5 and 6, where z A is respectively unconditioned and conditioned on z P .", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_624", "sentence1": "For Gradient Descent, we make use of this property by showing divergence increases as the strategies move from one pure strategy to another.", "sentence2": "strategies will never reach the boundary for some variants of FTRL.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_625", "sentence1": "These algorithms again have interesting properties in zero-sum games.", "sentence2": "in two-by-two zero-sum games, both steps (1) and (3) trivially extend for other variants of FTRL.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_626", "sentence1": "The time-average strategy converges to a O(η)-approximate Nash equilibrium (Cesa-Bianchi and Lugoisi [2006]).", "sentence2": "bailey and Piliouras [2018] show that the day-to-day behavior diverges away from interior Nash equilibria.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_627", "sentence1": "Despite the well established nature of these results recent work has revealed some surprising insights that come to challenge the traditional ways of thinking in this area.", "sentence2": "in the case of zero-sum games what is referred to as \"convergence\" to equilibrium, is the fact that when both agent apply regret-minimizing algorithms, both the time-average of the mixed strategy profiles as well as the utilities of the agents converge approximately to their Nash equilibrium values, where the approximation error can become arbitrarily close to zero by choosing a sufficiently small step-size.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_628", "sentence1": "While our approach draws high-level inspiration from MTL, we highlight key differences: whereas tasks are disjoint in MTL, slice tasks are formulated as micro-tasks that are direct extensions of a base task-they are designed specifically to learn deviations from the base-task representation.", "sentence2": "hPS: In the style of multi-task learning, we model slices as separate task heads with a shared backbone trained via hard parameter sharing.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_629", "sentence1": "Intuitively, if the model knows \"where\" in the 2-dim data space an example lives (as defined by SFs), it can condition on slice-specific features as it makes a final, slice-aware prediction.", "sentence2": "we consider two text-based relation extraction datasets: Chemical-Disease Relations (CDR), [41], in which we identify causal links between chemical and disease entities in a dataset of PubMed abstracts, and Spouses [9], in which we identify mentions of spousal relationships using preprocessed pairs of person mentions from news articles (via Spacy [17]).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_630", "sentence1": "A related line of work [KLS09, ABL17, DKK + 16, LRV16, DKK + 17, DKK + 18, DKS18, KKM18, DKS19, DKK + 19] has given polynomial time robust estimators for a range of learning tasks.", "sentence2": "[KLS09, ABL17, DKS18, DKK + 19] obtained efficient PAC learning algorithms for halfspaces with malicious noise [Val85,KL93], under the assumption that the uncorrupted data comes from a \"tame\" distribution, e.g., Gaussian or isotropic log-concave.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_631", "sentence1": "Instead of trying to guess the projection of w * onto the space of large eigenvectors all at once, we will do so in stages.", "sentence2": "we study the problem of properly learning large margin halfspaces in the agnostic PAC model.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_632", "sentence1": "We note that the inapproximability ratio of [DLS14] is close to being tight for a natural, yet restricted, family of improper learners.", "sentence2": "our proper hardness result holds against all proper learners under a widely believed worst-case complexity assumption.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_633", "sentence1": "Our algorithm tries to simulate the above described procedure by making appropriate guesses.", "sentence2": "we start by guessing a sequence of positive integers k (i) whose sum is at most 8/(δγ2).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_634", "sentence1": "We give learning algorithms and computational hardness results for this problem, for all values of the approximation ratio α ≥ 1, that are nearly-matching for a range of parameters.", "sentence2": "for the natural setting that α is any constant bigger than one, we provide an essentially tight complexity characterization.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_635", "sentence1": "We borrow an idea from [BFKV96] that in some sense allows us to \"reduce\" the general case to the large margin case.", "sentence2": "[BFKV96] (see also [DV04a]) developed a pre-processing routine that slightly modifies the distribution on the unlabeled points and guarantees the following weak margin property: After preprocessing, there exists an explicit margin parameter σ = Ω(1/poly(d, b)), such that any hyperplane through the origin has at least a non-trivial mass of the distribution at distance at least σ from it.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_636", "sentence1": "Moreover, this region is efficiently identifiable by a simple thresholding rule.", "sentence2": "we show that there exists a threshold T > 0 (which can be found algorithmically) such that the hypothesis sign(hwb , xi) has error bounded by η + \u000f in the region RT = {x : |hwb , xi| ≥ T}.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_637", "sentence1": "In the RCN model, the large margin case is easy because the learning problem is essentially convex.", "sentence2": "there is a convex surrogate that allows us to formulate the problem as a convex program.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_638", "sentence1": "As a corollary of our main result (Theorem 1.2), we answer this question in the affirmative.", "sentence2": "we obtain an efficient algorithm that achieves misclassification error arbitrarily close to η for all LTFs.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_639", "sentence1": "Our algorithm circumvents this difficulty by approaching the problem indirectly to find a nonproper classifier.", "sentence2": "our algorithm works in multiple rounds, where within each round only points with high value of | w, x | are considered.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_640", "sentence1": "We study the problem of distribution-independent PAC learning of halfspaces in the presence of Massart noise.", "sentence2": "we are given a set of labeled examples (x, y) drawn from a distribution D on R d+1 such that the marginal distribution on the unlabeled points x is arbitrary and the labels y are generated by an unknown halfspace corrupted with Massart noise at noise rate η < 1/2.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_641", "sentence1": "In fact one might not have access to the true distribution of the mapping, but just to the numerical outputs.", "sentence2": "one needs to consider more sophisticated metrics/divergences, such as the total variation distance The total variation distance is one of the most broadly used probability metrics.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_642", "sentence1": "Overconfidence alone decreases monotonically as we increase ↵ as shown in Figure 2i.", "sentence2": "we note that this approach is similar to the recent work described in [11] that utilizes mixup for improving sentence classification which is among the few works, besides ours, studying the effects of mixup in the NLP domain.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_643", "sentence1": "Both the confidence (captured by the winning score) as well as accuracy start out low and gradually increase as the network learns.", "sentence2": "what is interestingand concerning -is that the confidence always leads accuracy in the later stages of training; accuracy saturates while confidence continues to increase resulting in a very sharply peaked distribution of winning scores and an overconfident model.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_644", "sentence1": "This definition of SB-disentangled representations does not make any assumptions on what form the group action should take when acting on the relevant disentangled vector subspace.", "sentence2": "many subsequent tasks may benefit from a SB-disentangled representation where the group actions transform their corresponding disentangled subspace linearly.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_645", "sentence1": "One may utilize the measure to quantify better models [29] or directly use it as an objective function to optimize [37,1].", "sentence2": "measuring the discrepancy of GANs (and cGANs) is another challenging problem, since the data distribution remains unknown and the distribution GANs learn is implicit [35].", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_646", "sentence1": "We observe that the random samples (without rejection) often contain low-quality samples with uncertain and/or wrong classes.", "sentence2": "samples with high marginal values improve the quality (or vividness), and samples with high conditional values improve the class accuracy (but loses the diversity).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_647", "sentence1": "While both of the two approaches above, (a) and (b), can be used in a semi-supervised setting, cGANs of (b) provide a more natural framework for using both labeled and unlabeled data; 3 one can use the unlabeled data to learn p(x), and the labeled data to learn both p(x) and p(c|x).", "sentence2": "we focus on evaluating the second type of architectures, e.g., the auxiliary classifier GAN (ACGAN) [39].", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_648", "sentence1": "The generative adversarial network (GAN) [15] is arguably the most successful generative model in recent years, which have shown a remarkable progress across a broad range of applications, e.g., image synthesis [5,21,40], data augmentation [49,18] and style transfer [58,10,34].", "sentence2": "as its advanced variant, the conditional GANs (cGANs) [31] have gained a considerable attention due to its class-wise controllability [9,42,10] and superior quality for complex generation tasks [39,33,5].", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_649", "sentence1": "However, because the labels are biased, we must immediately assume that the corresponding accuracy measurements are also biased.", "sentence2": "we are crucially interested in evaluating accuracy on the unbiased ground truth labels, which are devoid of any such label bias.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_650", "sentence1": "It may be the case that label-bias is so obvious to most authors that it does not even occur to them to mention it; howbeit, the conspicuous absence of label-bias from papers on fairness perniciously pervades real-world discussions underlying the decisions about how to balance the trade-off between fairness and accuracy.", "sentence2": "we believe this finding to be of practical importance and worthy of highlighting.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_651", "sentence1": "We focused on demographic parity in this paper, but the ideas emphasized in this work, especially label bias, have potentially serious implications for other notions of fairness that go beyond even their relationship with accuracy.", "sentence2": "recent ways of assessing fairness such as disparate mistreatment, equal odds and equal oppurtunity involve error rates as measured against labeled data.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_652", "sentence1": "In this typical evaluation setting, if we train a set of classifiers that differ only in the extent to which their training objective functions enforce fairness, and then record their respective fairness and accuracy scores on a test set with such label bias, we see that increased fairness is achieved at the expense of accuracy (Figure 3a).", "sentence2": "because the labels are biased, we must immediately assume that the corresponding accuracy measurements are also biased.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_653", "sentence1": "While uncommon, some papers do indeed mention label-bias, including recent work that considers the largely hypothetical case: if we have access to unbiased labels, then we can propose a better way of evaluating fairness with \"disparate mistreatment\" [20].", "sentence2": "their emphasis is on new fairness metrics, not on its tradeoff with accuracy.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_654", "sentence1": "Critically, SSE progresses by stochastically transitioning between embeddings as opposed to a more brute-force regularization such as graph-based Laplacian regularization and ridge regularization.", "sentence2": "sSE integrates seamlessly with existing stochastic optimization methods and the resulting regularization is data-driven.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_655", "sentence1": "In stochastic optimization, we can replace the loss gradient for one movie's embedding with the other similar movie's embedding, and this will not significantly bias the gradient if the prior belief is accurate.", "sentence2": "if this exchange is stochastic, then it will act to smooth the gradient steps in the long run, thus regularizing the gradient updates.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_656", "sentence1": "We summarize our results in Table 5 and find that SSE-SE helps improving accuracy and BLEU scores on both dev and test sets in 10 out of 11 years from 2008 to 2018.", "sentence2": "on the last 5 years' test sets from 2014 to 2018, the transformer model with SSE-SE improves BLEU scores by 0.92 on average when compared to the baseline model without SSE-SE.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_657", "sentence1": "We summarize our results in Table 5 and find that SSE-SE helps improving accuracy and BLEU scores on both dev and test sets in 10 out of 11 years from 2008 to 2018.", "sentence2": "in [3], the authors masked 15% of words and 10% of the time replaced the [mask] token with a random token.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_658", "sentence1": "We continue to pre-train Google pre-trained BERT model on our crawled IMDB movie reviews with and without SSE-SE and compare downstream tasks performances.", "sentence2": "sSE integrates seamlessly with existing stochastic optimization methods and the resulting regularization is data-driven.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_659", "sentence1": "Our analysis then consists of two main parts where each part bounds one of the terms in the RHS above.", "sentence2": "if we only want to achieve zero inversions with high probability, how large does m need to be?", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_660", "sentence1": "Our analysis then consists of two main parts where each part bounds one of the terms in the RHS above.", "sentence2": "we first prove that given the partial comparisons seen so far, we can obtain a relatively good estimation to the rank of the arrived element, and then in the second part, we show that we can typically find an unassigned position in the close proximity of this estimated rank to assign to it.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_661", "sentence1": "Our algorithm maintains an ordering of the elements seen so far.", "sentence2": "the remainder of the paper is organized as follows.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_662", "sentence1": "Two important applications of this problem are search engines for document retrieval [L + 09, RJ05, LXQ + 07, CXL + 06, XL07] and collaborative filtering approaches to recommender systems [SLH10, SKB + 12, LY08, WRdVR08].", "sentence2": "our main result is a matching upper and lower bound for the secretary ranking problem.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_663", "sentence1": "Consider a university department that would like to assign the best scholarships available to the best students.", "sentence2": "scholarships arrive one at a time and the school needs to decide which student is assigned that scholarship knowing only the relative quality of the scholarships arrived so far.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_664", "sentence1": "Our algorithm maintains an ordering of the elements seen so far.", "sentence2": "when a new element arrives, the algorithm can find the position of this new element in the current ordering by binary search.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_665", "sentence1": "The total number of inversions can be approximated within a factor of 2 by the Spearman's footrule.", "sentence2": "we can write the cost of Algorithm 1 (up to a factor 2) as follows: This basically breaks the cost of the algorithm in two parts: one is the cost incurred by the estimation step and the other one is the cost of the assignment step.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_666", "sentence1": "In later time steps, we observe a large number of comparisons and using the randomness of elements arrival, the true rank of the elements can be estimated well.", "sentence2": "the main difficulty is that at these time steps many of the positions have already been assigned to some element arrived earlier and are hence not available.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_667", "sentence1": "Differently, HER opens up a new way to learn more from failures, assigning hindsight credit to primal experiences.", "sentence2": "it is limited by only applicable when combined with off-policy algorithms [3].", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_668", "sentence1": "A straightforward way to utilize self-imitate learning is to adopt the inverse dynamics.", "sentence2": "in most cases the actions stored in inverse dynamics are irrelevant to the goals.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_669", "sentence1": "PCHID greatly improves the learning efficiency of PPO.", "sentence2": "in general the efficiency of finding a transition sequence that is i-step solvable decreases as i increases.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_670", "sentence1": "In practice the k-step solvability can be treated as an evolving concept that changes gradually as the learning goes.", "sentence2": "at the beginning, an agent can only walk with small paces as it has learned from experiences collected by random movements.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_671", "sentence1": "However, the problems and models they study are very different.", "sentence2": "they assume that a decision-maker shows a subset of products to a user.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_672", "sentence1": "(2) Implementation: Our theoretical analysis does not require that each matrix estimation phase in REAL-Bandit is solved to completion.", "sentence2": "rEAL-Bandit does not need finding a global minimum of the relevant estimator's optimization (penalized maximum likelihood) problem.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_673", "sentence1": "However, existing matrix estimation results provide a bound on the estimation error of the whole matrix, which would be a crude upper bound for the estimation error of each single row.", "sentence2": "rEAL-Bandit includes a subroutine (called Row Enhancement) that refines the estimates in order to establish stronger row-wise guarantees that may be of independent interest (see §3 for details).", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_674", "sentence1": "The aim of this paper is to reduce this dependence under a low-rank assumption on the k × d matrix of parameters of the reward functions.", "sentence2": "each of the k reward functions is represented by a d-vector of coefficients (one coefficient per covariate) that is a row of the parameter matrix.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_675", "sentence1": "For example, the model in [4] assumed that the conditional intensity function exponentially decreases or increases with the elapsed time from the most recent event until the next event.", "sentence2": "using such an assumption can limit the expressive ability of the model and potentially deteriorate the predictive skill if the employed assumption is incorrect.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_676", "sentence1": "The conditional intensity function is then modeled as a function of the hidden state of the RNN.", "sentence2": "the RNN based models outperform the parametric models in prediction performance.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_677", "sentence1": "RNN based models usually assume a specific functional form for the time course of the intensity function of a point process (e.g., exponentially decreasing or increasing with the time since the most recent event).", "sentence2": "such an assumption can restrict the expressive power of the model.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_678", "sentence1": "In many cases, the occurrences of the event are correlated to each other in a certain manner, and information on future events may be extracted from the information of past events.", "sentence2": "the appropriate modeling of the dependence of the event occurrence on the history of past events is important for understanding the system and predicting future events.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_679", "sentence1": "In contrast to these studies, the log-likelihood function of our general model can be exactly evaluated without any numerical approximations because the integral of the hazard function is modeled by a feedforward neural network in our approach.", "sentence2": "a more accurate estimate can be efficiently obtained by our approach.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_680", "sentence1": "To evaluate the predictive performances based on a metric other than the log-likelihood, we also carry out the time prediction experiments.", "sentence2": "we use the median of the predictive distri-bution to predict the timing of the coming event and evaluate the prediction by the mean absolute error.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_681", "sentence1": "However, computing the assignments requires access to the optimal representation, which is not available.", "sentence2": "we propose an optimization scheme that alternates between updating the representation by minimizing our proposed loss given the current assignments of points to ground-truth representatives and updating the assignments given the current representation.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_682", "sentence1": "Moreover, the nature of our conditions, as opposed to asymptotic results, allowed to design the loss in (7).", "sentence2": "we use Euclidean distance for pairwise dissimilarities.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_683", "sentence1": "Notice that the F1 score is relatively stable with respect to the hyperparameter change.", "sentence2": "changing λ from 0.001 to 0.1 the performance over the dataset changes by at most 1.2% in F1 score, while changing ρinter and ρintra from 0.01 to 10, the performance changes by at most 0.6% and 2.1%, respectively.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_684", "sentence1": "This has the following effect: when a cluster has a small number of points, it could be considered as under-sampled, hence, to generalize better to test data, we need to have a better separation from other clusters, i.e., larger margin.", "sentence2": "for a cluster with a large number of points, the margin could be smaller as the chance of changing the distances between and within clusters by adding more samples to it would be low.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_685", "sentence1": "To address the problem, [50,52] try to learn a combination of different criteria, i.e., weights of a mixture of submodular functions.", "sentence2": "chengguang Xu would like to thank Dat Huynh and Zwe Naing for their help and advice with some of the implementations during his research assistantship at MCADS lab, which resulted in this work.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_686", "sentence1": "We draw data from two distribution families, the Normal family with mean a and variance b 2 , and the log-Normal family, with log-mean a log and log-variance b 2 log , under multiple parameter settings.", "sentence2": "we consider the impact of shifting the distribution location over [−40.0, 40.0], with small and large variance settings.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_687", "sentence1": "While this approach lets us deal with unbounded losses, naturally the statistical error guarantees are only as good as the confidence intervals available for the empirical mean deviations.", "sentence2": "strong assumptions on all of the moments of the loss are essentially unavoidable using the traditional tools espoused by Bégin et al. [3], which means the “heavy-tailed” regime cannot be handled, where all we assume is that a few higher-order moments are finite (say finite variance and/or finite kurtosis). ", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_688", "sentence1": "For most gradient-based NAS methods, an over-parameterized super-network is constructed firstly with all candidates paths included and one superior path is selected on each edge with the other candidates removed.", "sentence2": "signals in network often contain numerous channels during forward propagation, which means that a path is not the minimum separable structure unit in network and path-level search methods [3,24,22] limit the granularity of architecture search.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_689", "sentence1": "Recently, Neural Architecture Search has achieved great success in large-scale image classification.", "sentence2": "there have been limited works focusing on architecture search for object detection, mainly because the costly ImageNet pretraining is always required for detectors.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_690", "sentence1": "Qualitatively, we found that the estimates of feature importance provided by CXPlain were more focused on the subjectively more important semantic regions of the sample images from both MNIST and ImageNet (Figures 4 and 5; more in Appendix D).", "sentence2": "providing fast and accurate estimates of feature importance for high-dimensional data, and quantifying the uncertainty of such estimates remain open challenges.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_691", "sentence1": "Absent any prior knowledge about the structure of the input data, multilayer perceptrons (MLPs) are likely a sensible default choice.", "sentence2": "since architectures that exploit the spatial or temporal structure of input data have been shown to be efficacious, we reason that, depending on the data modality of the input features of the model to be explained, special-purpose architectures, such as convolutional neural networks [42] for images and attentive neural networks for texts [43], could perform better than MLPs.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_692", "sentence1": "Tree- [25][26][27] and rule-based [28] models have been used as mimic models.", "sentence2": "mimic models are not guaranteed to match the behavior of the original model.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_693", "sentence1": "To address this apparent dichotomy between performance and interpretability [6], researchers have developed a number of attribution methods that provide estimates of the importance of input features towards a model's output for specific types of models [4,[7][8][9][10][11][12][13][14][15], and for any machine-learning model [6,16].", "sentence2": "providing fast and accurate feature importance estimates for any machine-learning model is challenging because there exists a wide variety of intricate machine-learning models with different underlying model structures, algorithms, and decision functions, which makes it difficult to develop an optimised and unified approach to importance attribution.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_694", "sentence1": "Despite the population differences, however, we observe \"days_visited_vrs_pre\" continues to have a very significant positive association.", "sentence2": "we consider estimation of heterogeneous treatment effects with respect to a set of features X, of an endogenous treatment T on an outcome Y with an instrument Z.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_695", "sentence1": "For the UVFA network, it only needs to remember experiences in a local neighborhood.", "sentence2": "the training procedure requires much lower sample complexity.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_696", "sentence1": "One can pre-train the HER agent and then build map for planner.", "sentence2": "as an off-policy algorithm, HER can work with arbitrary exploration policy.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_697", "sentence1": "In their methods, the agent has to learn the high-level policy as another RL problem.", "sentence2": "we exploit the structure of our universal goal reaching problem and find the high-level policy by solving a pairwise shortest path problem in a small-scale graph, thus more data-efficient.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_698", "sentence1": "Given a predefined map of known nodes, edges, and weights, it runs the value iteration algorithm by ingeniously simulating the process through a convolutional neural network [19].", "sentence2": "we construct the map based upon the learned local model.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_699", "sentence1": "A UVFA learns to predict the cumulative rewards between all state-goal pairs.", "sentence2": "empirically, the value function for long-range goals is always hard to estimate and may consequently result in failed policy.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_700", "sentence1": "First, it takes longer time to sample enough state-goal pairs.", "sentence2": "at the early stage, only few state-goal samples have been collected, so learning from them requires heavy extrapolation by networks, which is well known to be unreliable.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_701", "sentence1": "For the UVFA network, it only needs to remember experiences in a local neighborhood.", "sentence2": "such map induces a new environment, where the action is to choose to move to another landmark.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_702", "sentence1": "To connect nearby landmarks, it leveraged a physical engine, which depends on sophisticated domain knowledge and limits its usage to other general RL tasks.", "sentence2": "for a given goal, the agent can receive non-trivial rewards only when it can reach the goal.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_703", "sentence1": "Interestingly, both conditions can be met with designs that involve the Lyapunov function V itself.", "sentence2": "the methods proposed in this paper can be understood as variable-stepsize discretizations, which are a popular class of methods in numerical analysis.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_704", "sentence1": "The real-time implementation of the closed-loop system can be tackled by considering a sample-and-hold implementation of (1) of the form with p(0) = p, where p is a sampled version of the state p. The most common approach consists of periodically sampling the state, selecting a stepsize small enough to ensure that the function V remains monotonically decreasing for the resulting system.", "sentence2": "the result in Theorem 3.3 links the convergence rate of the discrete-time algorithm to the Lyapunov decay and the stepsize of the state-triggered implementation of the continuous-time dynamics.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_705", "sentence1": "Their runtime is limited by the necessary Ω(1/γ) degree for polynomial approximation of the sign function shown by Eremenko and Yuditskii [9].", "sentence2": "to obtain nearly linear runtime, a new insight is required.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_706", "sentence1": "To improve upon this bound, we apply a similar approach of working with the convolution of f with a Gaussian.", "sentence2": "instead of applying standard stochastic AGD we consider accelerated methods which build a more sophisticated model of the convolved function in parallel.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_707", "sentence1": "Since the convolved function can be accessed efficiently in parallel by random sampling, working with the convolved function is comparable to working with the original function in terms of query depth (up to the sampling error).", "sentence2": "the paper achieves its depth bound by trading off the error induced by convolution with the depth improvements gained from stochastic variants of AGD.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_708", "sentence1": "Note that for any fixed query x, the probability (with respect to the random draw of v i ) that x is in C i is polynomially small in d. We now define the wall W as follows: it is equal to the function h outside of the correlation cones and the ball of radius , and it is extended by convexity to the rest of the unit ball.", "sentence2": "near-optimal convergence for such methods under Lipschitz bounds on the k-th derivatives were recently given by [Gasnikov et al., 2018 (and follow from our framework).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_709", "sentence1": "In this section, we establish the consistency of Interaction Hard Thresholding, in the standard setting where sparse recovery is established.", "sentence2": "this is our main result, described in Theorem 4.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_710", "sentence1": "Notice that IHT, if used for our quadratic problem, still suffers from quadratic space, similar to other techniques, e.g., the Lasso, basis pursuit, least angle regression [29,6,8].", "sentence2": "[19] recently considers a variant of IHT, where for each sample, only a random subset of features is observed.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_711", "sentence1": "In this section, we establish the consistency of Interaction Hard Thresholding, in the standard setting where sparse recovery is established.", "sentence2": "we establish convergence results under deterministic assumptions on the data and function, including restricted strong convexity (RSC) and smoothness (RSM).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_712", "sentence1": "Regression with interaction terms has been studied in the statistics community.", "sentence2": "many existing results consider under the assumption of strong/weak hierarchical (SH/WH) structure: the coefficient of the interaction term x j1 x j2 is non-zero only when both coefficients of x j1 and x j2 are (or at least one of them is) non-zero.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_713", "sentence1": "The inclusion of such higher-order \"interaction terms\" in regression often provides an easy way to increase accuracy in already-high-dimensional problems.", "sentence2": "this explodes the problem dimension from linear O(p) to quadratic O(p 2 ), and it is common to look for sparse interactions (typically via heuristics).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_714", "sentence1": "If one wants to make an existing domain adaptation algorithm private with respect to any pair of participants, one has to add noise to gradients computed at every step.", "sentence2": "any entity can compute the marginal gain of a new point y by Thus, the curator needs to only share ) and h 2 is differentially private with respect to h 1 (x i ).", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_715", "sentence1": "Thus, it is sufficient to have a significantly smaller number of iterations than that in Theorem 4.", "sentence2": "t sub is set to be small.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_716", "sentence1": "Further, over the course of multiple releases, any participant must not be able to acquire any information about previous data points of other participants.", "sentence2": "the key issue is that the releases of the curator must be differentially private while enabling the computation of the (non-linear) marginal gain, over all the iterations of the protocol.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_717", "sentence1": "Small incremental changes to the empirical distribution need to be matched incrementally.", "sentence2": "it is sufficient to have a significantly smaller number of iterations than that in Theorem 4.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_718", "sentence1": "The authors used parameterized Hamiltonian flows for distribution embeddings, which limits its scalability and expressiveness.", "sentence2": "dDE fails if the search space does not contain the target distribution, while our formulation only requires the support of the proposal distribution to cover that of the target.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_719", "sentence1": "More recently, dynamic dual embedding (DDE) explored a primal-dual view of MLE [15,16], while Stein implicit learning (SIL) [46,41] and kernel score estimation [60] match the landscape of the potential with that of kernel-smoothed empirical observations.", "sentence2": "these approaches are susceptible to poor scalability (SM, MCMC-MLE), biased estimation (CD), and computational (DDE, SIL) and statistical (NCE) efficiency issues.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_720", "sentence1": "Our next experiment considers FML-based training for latent variable models and generative modeling tasks.", "sentence2": "we directly benchmark FML against the VAE [37], for modeling complex distributions, such as images and natural language, for real-world applications.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_721", "sentence1": "In practice, this integral is approximated with averaging over a finite number of Monte Carlo samples.", "sentence2": "using the existing finite-sample Monte Carlo estimate of Zθ¸ will lead to a biased approximation of the log-likelihood objective (see Section 2.1).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_722", "sentence1": "It is easy to show a similar convergence rate for an adaptation of this algorithm to our setting (see Appendix C), but we stick to our present algorithm because of its simplicity.", "sentence2": "postShift performs the best on the fairness metric, but on all datasets except COMPAS, fairs poorly on the constraint.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_723", "sentence1": "We consider the fairness goal of training a classifier that yields at least as high a F-measure for the protected group as it does for the rest of the population, and impose this as a constraint.", "sentence2": "we seek to maximize the overall F-measure subject to the constraint: Fmeasure prt Fmeasure other 0.02.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_724", "sentence1": "Our approach generalizes many existing algorithms (see Table 2), and makes possible new algorithms with more flexibility and tighter handling of non-linear rate constraints.", "sentence2": "we give a new method (Algorithm 2) that can handle a wider range of performance metrics than previous surrogate methods (such as e.g. KL-divergence based metrics that only take inputs from a restricted range), and can be applied to constrained training problems without the risk of over-constraining the model because it needs to use surrogates less. ", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_725", "sentence1": "[33] handle constraints that are sums-of-ratios, but do so by solving a large number of linearly constrained sub-problems, with the number of sub-problems growing exponentially with the number of rates.", "sentence2": "each player can do Best Response (BR), Online Gradient Descent (OGD) or Follow-The-Leader (FTL), and the game is zero-sum (ZS) or not.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_726", "sentence1": "Several papers [21][22][23] have made initial progress in defining scattering graph representation and studying their stability properties with respect to metric deformations of the domain.", "sentence2": "most of these results offer bounds that depend on the graph topology and do not hold for certain graphs or when graphs are very large.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_727", "sentence1": "Regular scattering transforms have been proven invariant to translations and stable to perturbations (or deformations) that are close to translations.", "sentence2": "the difference on the scattering transform of the original data and that of the perturbed data, is proportional to the size of the perturbation.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_728", "sentence1": "The effect of the pointwise nonlinearities is to cause a spillage of information throughout the frequency spectrum, in particular, into low-eigenvalue frequencies, which can then be discriminated in a stable fashion.", "sentence2": "gSTs are stable and discriminative information processing architectures.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_729", "sentence1": "That is, by changing the eigenvalues λi on which the wavelet h(λ) is instantiated, the filter taps hi are changed, and so does the output λi in virtue of (6).", "sentence2": "the first necessary result is to quantify the change in the output of a wavelet filter.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_730", "sentence1": "We begin by formulating an optimization problem whose solution is a form of optimal representation.", "sentence2": "we seek a state representation from which we can best approximate the value function of any stationary policy for a given Markov Decision Process.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_731", "sentence1": "While a natural choice is to look for policies that maximize representation error, this poses the problem of how to parametrize the policies themselves.", "sentence2": "a policy parametrized using the representation φ may not provide a sufficient degree of \"adversariality\".", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_732", "sentence1": "So far we have argued that solving the RLP leads to a representation which is optimal in a meaningful sense.", "sentence2": "solving the RLP seems computationally intractable: there are an exponential number of deterministic policies to consider (Prop.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_733", "sentence1": "We contrast the AVF-driven representation with one learned by predicting the value function of random deterministic policies (RP).", "sentence2": "these policies are generated by assigning an action uniformly at random to each state.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_734", "sentence1": "The good empirical performance of distributional reinforcement learning (Bellemare et al., 2017) has also been attributed to representation learning effects, with recent visualizations supporting this claim (Such et al., 2019).", "sentence2": "while there is now conclusive empirical evidence of the usefulness of auxiliary tasks, their design and justification remain on the whole ad-hoc.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_735", "sentence1": "So far we have argued that solving the RLP leads to a representation which is optimal in a meaningful sense.", "sentence2": "we leverage this perspective to provide formal evidence regarding the usefulness of value functions as auxiliary tasks.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_736", "sentence1": "The seminal work by Wen and Van Roy [36] proposed an algorithm, optimistic constraint propagation (OCP), which enjoys polynomial sample complexity bounds for a family of Q-function classes, including the linear function class as a special case.", "sentence2": "their algorithm can only deal with deterministic systems, i.e., both transition dynamics and rewards are deterministic.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_737", "sentence1": "More recent papers provided refined analyses that exploit benign properties of the MDP, e.g., the gap between the optimal action and the rest [28,38], which our algorithm also utilizes.", "sentence2": "it is hard to generalize the exploration techniques in these previous works, since they all rely on the fact that the total number of states is finite.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_738", "sentence1": "The idea of using function approximation was proposed at least 60 years ago [27], where linear functions are used to approximate the value functions in playing checkers.", "sentence2": "even in the most basic setting, Q-learning with linear function approximation, there is no provably efficient algorithm in the general stochastic setting.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_739", "sentence1": "In practice, it is impossible to get access to the underlying distributions D 1 and D 2 .", "sentence2": "we use samples generated from these two distributions instead.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_740", "sentence1": "We further show this potential function is at always polynomially upper bounded by the size of the policy set.", "sentence2": "we can conclude the size of IIh is polynomially upper bounded.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_741", "sentence1": "This is because of approximations when learner is computing the policy via projection, which in turn leads to errors on the teacher side when approximating ΩL r (refer to discussion in Footnote 2).", "sentence2": "the numbers show the average reward over 10 randomly generated object-worlds.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_742", "sentence1": "The improvement on Sketch is notable because Sketch is the only colorless domain out of the four domains in PACS.", "sentence2": "when tested with the other three domains, a model may learn to exploit the color information, which is usually local, to predict, but when tested with Sketch domain, the model has to learn colorless concepts to make good predictions.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_743", "sentence1": "He is also grateful to Salesforce Research, Facebook Research, and Amazon AI for faculty awards supporting his lab's research on robust deep learning under distribution shift.", "sentence2": "our paper extends the setup of (Wang et al., 2019) and empirically studies the problem of developing image classifiers robust to a variety of natural shifts without leveraging any domain information at training or deployment time.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_744", "sentence1": "While we did not give a clear choice of which PAR to use, we note that none of the variants of PAR outperform the vanilla PAR consistently.", "sentence2": "the vanilla PAR outperforms most comparable baselines in the vast majority of our experiments.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_745", "sentence1": "In addition to the superior performances we achieved through these experiments, we expected to further challenge our method at real-world scale.", "sentence2": "we also constructed a dataset that matches the ImageNet classification validation set in classes and scales but contains only sketch-alike images.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_746", "sentence1": "The standard estimator for the distribution mean is the sample mean.", "sentence2": "for distributions with unbounded support, e.g., Gaussians, the global sensitivity of the sample mean is infinite.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_747", "sentence1": "We provide methods for scaling noise in an instance-dependent way and demonstrate that they provide greater accuracy under average-case distributional assumptions.", "sentence2": "we consider the basic problem of privately estimating the mean of a real distribution from i.i.d. samples.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_748", "sentence1": "Of course, the mean and median are closely related.", "sentence2": "there is a subtle -but important -difference: Whereas the standard deviation provides the appropriate scale for the accuracy of an estimate of the mean, the reciprocal of the probability density around the median provides the appropriate scale for an estimate of the median [31].", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_749", "sentence1": "If the stronger pure differential privacy guarantee is preferable, the Student's T distribution is likely best.", "sentence2": "this has no third moment and consequently heavy tails.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_750", "sentence1": "We provide methods for scaling noise in an instance-dependent way and demonstrate that they provide greater accuracy under average-case distributional assumptions.", "sentence2": "the mean and median (the extreme cases of the trimmed mean) have both been studied extensively in the differential privacy literature.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_751", "sentence1": "Feldman and Steinke [16] use a median-of-means algorithm to privately estimate the mean, yielding a guarantee similar to Theorem 8.", "sentence2": "their algorithm partitions the dataset into evenlysized subdatasets and computes the mean of each subdataset.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_752", "sentence1": "A recent line of work [15,5,26,4] has developed variants of differential privacy which permit tighter analyses of privacy loss over multiple releases of statistics as compared to both pure and approximate differential privacy.", "sentence2": "the notion of concentrated differential privacy (CDP) [15,5] has a simple and tight composition theorem for analyzing how privacy degrades over many releases while accommodating most of the key algorithms in the differential privacy literature, including addition of Gaussian noise calibrated to global sensitivity.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_753", "sentence1": "The method makes use of two Q-functions to overcome the positive bias incurred by overestimation of Q-value, which is known to yield a poor performance [15,9].", "sentence2": "these two Q-functions are parametrized by different parameters θi and are independently trained to minimize JQ(θi). ", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_754", "sentence1": "An MDP is defined by a tuple (S, A, P, r, P 0 , γ), where S is the state space and A the action space with |A| actions.", "sentence2": "we have shown there are many regularization functions that can lead to a sparse but multi-modal optimal policy such as trigonometric and exponential functions.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_755", "sentence1": "In practice, we apply neural networks to parameterize the Q-value and policy to increase expressive power.", "sentence2": "we model the regularized Q-value function Qθ(s, a) and a tractable policy πψ(a|s).", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_756", "sentence1": "Moreover, under our framework, many regularization terms can bring multi-modality and sparsity, which are potentially useful in reinforcement learning.", "sentence2": "we present sufficient and necessary conditions that induce a sparse optimal policy.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_757", "sentence1": "When a suggested routine is unfortunately congested, an alternative routine could be provided by a multi-modal policy, which can't be provided by a deterministic policy without evoking a new computation.", "sentence2": "in a real-life application, we hope the optimal policy to possess thee property of multi-modality.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_758", "sentence1": "In practice, we apply neural networks to parameterize the Q-value and policy to increase expressive power.", "sentence2": "we have presented the logical and mathematical foundations of these properties and also conducted the experimental results.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_759", "sentence1": "in Fig. 2 (b), we analyze the functional changes over time.", "sentence2": "we fit the linear mapping function between neuron's firing rate and target trajectory in every 20-second temporal window with a stride of 1 second, and illustrate the distribution of the slope parameter.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_760", "sentence1": "Lastly, since the stratified structure contains all of the examples, it is managed mostly on disk, with a small in-memory buffer to speed up I/O operations.", "sentence2": "this performance requires that the memory size is sufficient to hold a 2-3 multiple of the training set size.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_761", "sentence1": "We distinguish between two types of average potentials: the expected potential or true potential: and the average potential or empirical potential: The ultimate goal of the boosting algorithm is to minimize the expected potential, which determines the true error rate.", "sentence2": "most boosting algorithms, including XGBoost and LightGBM, focus on minimizing the empirical potential b (S T ), and rely on the limited capacity of the weak rules to guarantee that the true potential is also small.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_762", "sentence1": "The value of n eff will drop to about 80.", "sentence2": "it is achieved with no penalty in accuracy, and with a speedup of 10-100 over XGBoost in disk mode.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_763", "sentence1": "The agent's uncertainty over the environment is represented as a prior distribution over θ.", "sentence2": "θ¸ will be treated as a random variable in the agent's mind.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_764", "sentence1": "As we will see in the following two lemmas, the structure of MDPs allows us to break down the deviation of value functions and the information gain at the level of state-action pairs.", "sentence2": "it would be sufficient to construct confidence sets for the reward and transition functions for individual state-action pairs.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_765", "sentence1": "By Lemma 4, since noise is additive, it is sufficient to construct confidence sets on Y a = θ a. Lemma 5.", "sentence2": "if (1) is satisfied with reasonable values for Γ and , a large expected regret on the left-hand side would imply that the right-hand side must be large as well, meaning that the agent should gain a lot of information about the environment.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_766", "sentence1": "Hoeffding's inequality [10] or empirical Bernstein's inequality [14], to find a valid threshold hα(yn).", "sentence2": "they are not data-dependent and only use the tail information, rather than fully exploit the whole distribution knowledge.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_767", "sentence1": "We use similar notations for Py(·), Ey(·)  with respect to y only.", "sentence2": "they are not data-dependent and only use the tail information, rather than fully exploit the whole distribution knowledge.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_768", "sentence1": "Hoeffding's inequality [10] or empirical Bernstein's inequality [14], to find a valid threshold hα(yn).", "sentence2": "details are deferred to Section B.3 in the supplement.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_769", "sentence1": "As mentioned in [36], there are two disadvantages for median-of-means approach: (a) it involves an additional tuning parameter; (b) it is numerically unstable for small sample size.", "sentence2": "we identify a class of heavy-tailed bandits (sub-Weibull bandit) where mean estimators can still achieve regret bounds of the same order as those under sub-Gaussian reward distributions.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_770", "sentence1": "Existing techniques for constructing confidence bounds are typically built upon various concentration inequalities, which thus lead to over-exploration.", "sentence2": "the proof relies on a precise characterization of p-th moment of a Weibull random variable and standard symmetrization arguments.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_771", "sentence1": "Throughout the paper, we assume that the columns of the data matrix have `2-norm at most one 1 .", "sentence2": "in our analysis of our differentially private covariance estimation mechanism, we will focus on bounding the Frobenius error.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_772", "sentence1": "In this section, we show that a private estimate of the covariance matrix C = XX > summarizes the data sufficiently well for all of these ridge regression learning tasks with only a one-time privacy cost.", "sentence2": "we can view differentially private covariance estimation as a database sanitization scheme for ridge regression.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_773", "sentence1": "There was also an attempt by Jiang et al. [2016] to use Wishart-distributed noise to privately estimate a covariance matrix.", "sentence2": "imtiaz and Sarwate [2016] proposed the same algorithm and later discovered that the algorithm was in fact not differentially private.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_774", "sentence1": "Recall that the Gaussian mechanism has an additional failure probability , thus the privacy guarantees we obtain are strictly better for the same value of ✏.", "sentence2": "it is particularly striking that we consistently beat the Gaussian mechanism even for the very relaxed value of = .001.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_775", "sentence1": "We tried fixing the generator weights and optimizing the noise so that it generates the target image, under the assumption that sightly modifying the optimized noise would produce a variant of the original.", "sentence2": "naively implementing this idea with BigGAN did not reconstruct the image well, as shown in the sample in Fig.  1(b).", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_776", "sentence1": "As shown in Figure 1(g), we divide the images into a 3 X 3 grid and linearly combine the cells with the weights w produced by a CNN conditioned on the two images.", "sentence2": "where is element-wise multiplication, and w is resized to the image size keeping the block structure.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_777", "sentence1": "Another way to address data scarcity is to synthesize additional training examples, for example by using off-the-shelf Generative Adversarial Networks (GANs) [3,13].", "sentence2": "classifiers trained from GAN-generated images are typically inferior to those trained with real images, possibly because the distribution of generated images may be biased towards frequent patterns (modes) of the original image distribution [26].", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_778", "sentence1": "Since only a few training images are available in the target domain, only scale and shift parameters of the batch normalization of G are updated in practice.", "sentence2": "only the and of each batch normalization layer are updated in each layer, where x is the input feature from the previous layer, and E and Var indicate the mean and variance functions, respectively.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_779", "sentence1": "This is different from the neural network literature [45,25], where adversarial training usually provides better LRTE and significantly better test error than methods providing provable robustness guarantees.", "sentence2": "our upper bound on the robust loss is tight and tractable and thus adversarial training should not be used as it provides only a lower bound and minimization of an upper bound makes more sense than minimization of a lower bound.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_780", "sentence1": "We can solve this certification problem for the robust test error exactly and efficiently by noting that the objective and the attack model 1 (✏) is separable wrt the input dimensions.", "sentence2": "we have to solve up to d simple one-dimensional optimization problems.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_781", "sentence1": "We note that our methods are primarily suitable for tabular data, but in the literature on robustness of neural networks there are no established tabular datasets to compare to.", "sentence2": "we compare our robust boosted trees to the convolutional networks of [73,16,75,25,13] on MNIST, FMNIST, and CIFAR-10.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_782", "sentence1": "In terms of provable robustness (URTE), our method is competitive to many provable defenses for CNNs.", "sentence2": "we outperform the LP-relaxation approach of [73] on all three datasets both in terms of test error and upper bounds.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_783", "sentence1": "This cannot be guaranteed with robust splits without pruning since the tree construction process is greedy, and some training examples are also influenced by splits at different branches.", "sentence2": "in order to control the upper bound on the robust loss globally over the whole tree as in (8), and not just for the current subtree as in ( 9), we need a post-hoc approach that takes into account the structure of the whole tree.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_784", "sentence1": "In general, for an arbitrary loss L, there is no closed-form minimizer wrt w l and w r .", "sentence2": "we can minimize such an objective using, e.g. coordinate descent.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_785", "sentence1": "Thus, in order to control the upper bound on the robust loss globally over the whole tree as in (8), and not just for the current subtree as in ( 9), we need a post-hoc approach that takes into account the structure of the whole tree.", "sentence2": "we have to use pruning.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_786", "sentence1": "Note that it is possible to find a separating hyperplane between the origin and the collection of points {F ai } if and only if the problem is in Regime 3.", "sentence2": "the problem of inverse reinforcement learning can be viewed as a one class support vector machine (or as a two class support vector machine with the origin as the negative class) problem in this regime.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_787", "sentence1": "Methods such as Abbeel & Ng (2004) looked at using IRL to solve the apprenticeship learning problem by trying to find a reward function that maximizes the margin of the expert's policy.", "sentence2": "reinforcement Learning is the process of generating an optimal policy for a given Markov Decision Process (MDP) along with a reward function.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_788", "sentence1": "Various algorithms to solve the IRL problem have been proposed including linear programming Ng & Russel (2000), Hybrid IRL Neu & Szepesvári (2007), Maximum Margin Planning Ratliff et al.  (2006), Multiplicative Weights for Apprenticeship Learning Syed et al. (2008) and Bayesian estimation Ramachandran & Amir (2007).", "sentence2": "the problem of inverse reinforcement learning can be viewed as a one class support vector machine (or as a two class support vector machine with the origin as the negative class) problem in this regime.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_789", "sentence1": "In our test, however, we are fitting a MV-Kumaraswamy variational posterior.", "sentence2": "we compute gradient variance, for all methods, according to variance's more common definition.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_790", "sentence1": "We do not use any explicit regularization.", "sentence2": "figure 5: Variance of the ELBO's gradient's first dimension for GRG [22], RSVI [18], IRG [4], and MVK (ours) when fitting a variational posterior to Categorical data with 100 dimensions and a Dirichlet prior.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_791", "sentence1": "This example qualitatively illustrates corollary 1.", "sentence2": "as we demonstrate in section 2.2 and fig.  1, approximating sparse Dirichlet samples with the Kumaraswamy stick-breaking decomposition without accounting for the ordering dependence produces a large bias in the samples’ last dimension.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_792", "sentence1": "We also experimented with a non-variational autoencoder.", "sentence2": "in the majority of experiments, that network overfit to a 1D manifold.", "label": "contrasting", "domain": "NeurIPS"}
{"id": "test_793", "sentence1": "Overfitting the autoencoder to history will degrade feature quality on future scenarios.", "sentence2": "we employ early stopping to be conservative with autoencoder training.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_794", "sentence1": "To remedy this problem, we maintain an experience replay buffer [Mnih et al., 2015] of snapshots from multiple simulations.", "sentence2": "this experiment also allows comparisons over controllers of similar size.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_795", "sentence1": "As output, weights for an observer function with a descriptive latent space are learned.", "sentence2": "recent breakthroughs have demonstrated capable computational methods for both controlling (Heess et al.  [2017], Schulman et al. [2017], Lillicrap et al. [2015]) and designing (Ha et al. [2017], Spielberg et al., Wampler and Popovic [2009]) rigid robots.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_796", "sentence1": "While we try to compare the same number of clusters and latent variables in experiments (since inputs from a cluster give highly dependent data) we acknowledge that each cluster (in 2D) provides six inputs to the controller.", "sentence2": "this experiment also allows comparisons over controllers of similar size.", "label": "reasoning", "domain": "NeurIPS"}
{"id": "test_797", "sentence1": "In this paper, we address the problem of learning low-dimensional robot state while simultaneously optimizing robot control and/or material parameters.", "sentence2": "we require a representation applicable to physical control of real-world robots.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_798", "sentence1": "As output, weights for an observer function with a descriptive latent space are learned.", "sentence2": "we learn a variational autoencoder Welling, 2013, Rezende et al., 2014] that takes, as input, a state description of a robot and minimizes the reconstruction cost of said state.", "label": "entailment", "domain": "NeurIPS"}
{"id": "test_799", "sentence1": "Using SWAG as a transfer task for COPA sees an 8 point improvement.", "sentence2": "gLUE provides a lightweight classification API and no restrictions on model architecture or parameter sharing, which seems to have been well-suited to recent work in this area.", "label": "neutral", "domain": "NeurIPS"}
{"id": "test_800", "sentence1": "From the traces of the mesh STAs, it has been observed that, in case of inter­ference between 20 MHz and 40 MHz channels (we call this as asymmetric channel interference ) at STA-2, the packets transmitted through the 40 MHz channel gets su.ered.", "sentence2": "sTA-2 can decode the over­heard control packets from STA-3, however fails to de­code the packets from STA-1 in case of an asymmetric channel interference.", "label": "reasoning", "domain": "Networks"}
{"id": "test_801", "sentence1": "Further, higher MCS levels require higher receiver sensitivity.", "sentence2": "mCS 13 at 20 MHz channel performs best with poor signal quality.", "label": "reasoning", "domain": "Networks"}
{"id": "test_802", "sentence1": "The nodes use one of the control channels to send this information ¯ 2.", "sentence2": "but none of the above schemes make use of network wide wavelength availability-states (which is more precise) to adapt the offset-time and hence cannot make full use of the advantages offered by JET protocol.", "label": "neutral", "domain": "Networks"}
{"id": "test_803", "sentence1": "The duration of each simulation run is 900 seconds.", "sentence2": "in a network with a high level of node mobility, links break very frequently.", "label": "neutral", "domain": "Networks"}
{"id": "test_804", "sentence1": "The simulation results show that TADL is practical in a stationary network or a network with a low to medium level of mobility.", "sentence2": "area-S is the smallest searching area that includes S and D.", "label": "neutral", "domain": "Networks"}
{"id": "test_805", "sentence1": "At the same time, these attacks reduce the traffic load in the network reducing the level of PDR drop caused by congestions.", "sentence2": "packet-dropping attacks can offset packet loss caused by network congestions.", "label": "entailment", "domain": "Networks"}
{"id": "test_806", "sentence1": "each node esti­mates a neighbouring node trust value based on the quality of the link connecting to the neighbouring node.", "sentence2": "the trust value assigned to node Ni actually reflects the reliability of the link linking this node and node Ni.", "label": "entailment", "domain": "Networks"}
{"id": "test_807", "sentence1": "An interesting observation in scenarios 1, 1a, 2 and 2a is that the cost increases almost linearly with the number of subscriptions upto a certain point (number of subscriptions = 4000).", "sentence2": "the average cost for matching a subscription is almost constant.", "label": "reasoning", "domain": "Networks"}
{"id": "test_808", "sentence1": "At time 2, when the new subscription f3 occurs, RMCEP-Online is invoked again, which in turn calls RMCEP-Aprx with the subscriptions f2 (since e is not yet placed to satisfy f2) and f3.", "sentence2": "in each round of the algorithm, the chunk with the minimum AOC is selected, and added to the .nal solu­tion.", "label": "neutral", "domain": "Networks"}
{"id": "test_809", "sentence1": "On the other hand, the number of RSUs do not have a significant direct impact on the service provider s cost, and hence the provider can install or rent a much smaller number of RSUs placed suitably around the city to achieve the same cost.", "sentence2": "note that sufficient number of RSUs are still needed to be suitably placed as all subscriptions and events are also reported to the service provider through the RSUs.", "label": "contrasting", "domain": "Networks"}
{"id": "test_810", "sentence1": "Its main aim is to provide an overall ordering of options.", "sentence2": "in each observation, the residual energy value is measured after transmitting 10 packets because for each packet transmission, very small amount of energy is consumed.", "label": "neutral", "domain": "Networks"}
{"id": "test_811", "sentence1": "While their results provide interesting insights into the relative impact as the load of the system changes from light-to-heavy load, the simulations also suggest that under light load correlated service times typically result in an increase in the end-to-end delays.", "sentence2": "we typically observe a decrease both under light and heavy load.", "label": "contrasting", "domain": "Networks"}
{"id": "test_812", "sentence1": "Secondly, the traditional context-aware access control policies may look fine-grained for traditional data files, but is not fine-grained enough for sensor data, especially because sensor data collection is context aware.", "sentence2": "the Portal app allows the smartphone user to review the sensor usage reports of all installed apps.", "label": "neutral", "domain": "Networks"}
{"id": "test_813", "sentence1": "How does a server acknowledge receipt of a range of data?", "sentence2": "even if our vision of any IoT device connecting to any smartphone proves too radical a departure from the status quo, the basic ideas could still be deployed in more constrained administrative domains, like a home, office, or university campus.", "label": "neutral", "domain": "Networks"}
{"id": "test_814", "sentence1": "For example, information about the location of the peripheral or the current global time may be difficult for an IoT device to obtain, but straightforward for a smartphone.", "sentence2": "how should gateways choose when to forward data?", "label": "neutral", "domain": "Networks"}
{"id": "test_815", "sentence1": "At what point can a peripheral be sure its data reached the end server?", "sentence2": "in this paper, we identify some of the key issues and begin to explore them, with the goal of raising awareness and generating discussion about both the opportunities and challenges.", "label": "neutral", "domain": "Networks"}
{"id": "test_816", "sentence1": " Unfortunately, these models are necessarily tied to specific network types and locations.", "sentence2": "we observe that high-fidelity information about the cellular channel is readily available from the radio-layer signaling protocols employed by high-speed cellular networks (e.g., HSPA+ and LTE).", "label": "contrasting", "domain": "Networks"}
{"id": "test_817", "sentence1": "In such a scenario, CQIC tries to match the the achievable bandwidth with the channel capacity offered by the cellular base station for the UE.", "sentence2": "the fairness among CQIC and TCP flows to different UEs are provided by the underlying scheduling algorithm implemented at the base station.", "label": "entailment", "domain": "Networks"}
{"id": "test_818", "sentence1": "We configure the UE in diagnostic mode, and the QXDM tool continuously queries and collects various radio and chip-­level information.", "sentence2": "we use the HS-DPCCH-INFO and HS-DECODE-STATUS log packets to retrieve the CQI and DTX information, respectively.", "label": "entailment", "domain": "Networks"}
{"id": "test_819", "sentence1": "We build upon Google's QUIC [15] framework, which is a new transport protocol based on UDP, to implement a CQIC prototype.", "sentence2": "we reuse the RTT estimation and reliable packet delivery modules in QUIC.", "label": "entailment", "domain": "Networks"}
{"id": "test_820", "sentence1": "NetRadar [2] collects traces from mobile devices to acquire localized link throughput statistics; OpenCellID [5] collects cell tower information and correlates them with user locations to establish network density and coverage maps; and finally, OpenSignal [12] measures the signal strength of cell towers to estimate achievable throughputs.", "sentence2": "none of the current approaches yet takes a holistic view along all operators and protocol levels to enable advanced modelling, e.g. through spatial statistics and machine learning based filtering methods, to build highly relevant coverage and performance estimates.", "label": "contrasting", "domain": "Networks"}
{"id": "test_821", "sentence1": "Intuitively, Nightcrawler should use the above collection of reflected TX beam directions to drive imag­ing.", "sentence2": "the TX focuses its transmissions on these directions (by rotating its beam repeatedly across them in a round-robin fash­ion) while the RX locates and images object(s) in each direction.", "label": "entailment", "domain": "Networks"}
{"id": "test_822", "sentence1": "For these reasons the academic community is concentrating its efforts in searching for mobile technologies that enable effectively the exchange of information between the vehicles.", "sentence2": "it would be possible to receive and pass information to other automobiles warning them about accidents, traffic jams and easy ways where they have passed, allowing other drivers to choose alternative paths through the messages received.", "label": "reasoning", "domain": "Networks"}
{"id": "test_823", "sentence1": "In the last decade, many techniques and tools for available bandwidth estimation have been developed.", "sentence2": "their accuracy still remains a challenge [24].", "label": "contrasting", "domain": "Networks"}
{"id": "test_824", "sentence1": "Normal has 23.9Mbps at 0m when the number of terminals is 10.", "sentence2": "if multiple STAs transmit frame to an AP at the same time and the collision occurs, the behavior depends on the difference of received power between two frames.", "label": "neutral", "domain": "Networks"}
{"id": "test_825", "sentence1": "This study uses Eq.(1) as the condition for activating MIMmode.", "sentence2": "fig.2 represents the model of the assumed environment.", "label": "neutral", "domain": "Networks"}
{"id": "test_826", "sentence1": "As mentioned in section 4.2, when the number of ter­minals is one, the parallel transmission case decreases be­cause the number of collision among WLANs decreases by coexisting EIFS and DIFS.", "sentence2": "if the number of ter­minals is larger than two, the number of collision in same WLAN decreases and the one among WLANs increases.", "label": "contrasting", "domain": "Networks"}
{"id": "test_827", "sentence1": "It is because the increment using binary increase cannot avoid the collision when CWmin is small.", "sentence2": "if CW = 1, CW becomes 2 by binary increase in case of Collision/Error.", "label": "entailment", "domain": "Networks"}
{"id": "test_828", "sentence1": "If the difference is larger than SINR, a frame with stronger received power results in Collision/Capture and the other frame with weak received power results in Collision/Error.", "sentence2": "if the difference is smaller than SINR, both frames result in Collision/Error.", "label": "contrasting", "domain": "Networks"}
{"id": "test_829", "sentence1": "It is because the distance between WLANs is comparatively large and the capture probability becomes higher.", "sentence2": "the collision results in Collision/Capture.", "label": "reasoning", "domain": "Networks"}
{"id": "test_830", "sentence1": "We use the PC (Toshiba dynabook SSRX2) as the sender terminal whose OS is ubuntu10.04 (the chipset and WLAN driver are AR928X and ath9k, respectively).", "sentence2": "when Collision/Error occurs, the proposal increases CWmin drastically.", "label": "neutral", "domain": "Networks"}
{"id": "test_831", "sentence1": "If the Capture Probability is high, the throughout improves by parallel transmission when the collisions often occur.", "sentence2": "the modified small CW causes Collision/Capture and Parallel Transmission.", "label": "reasoning", "domain": "Networks"}
{"id": "test_832", "sentence1": "One of the most distinctive and substantial characteristics of VANETs is non-infrastructure.", "sentence2": "there is no fixed routing, like access point (AP), or any central unit [8, 9].", "label": "entailment", "domain": "Networks"}
{"id": "test_833", "sentence1": "In the past, several schemes have been proposed to reduce the broadcast storm problem.", "sentence2": "they have been only validated using simple scenarios such as a highway (several lanes, without crossroads) [5,6], or with the help of road side units [7].", "label": "contrasting", "domain": "Networks"}
{"id": "test_834", "sentence1": "However, significant overheads in terms of communication and storage are incurred due to the flooding or multi-hop forwarding [10, 11]. 3) Without the network infrastructure, steady connection between nodes is difficult to be guaranteed, especially in large scale VANETs.", "sentence2": "the scalability is difficult to be achieved in large scale VANETs [12].", "label": "entailment", "domain": "Networks"}
{"id": "test_835", "sentence1": "If a node has more neighbors, the node resides in a more important position.", "sentence2": "the node with larger connectivity degree should have more probability to be elected as the CH.", "label": "reasoning", "domain": "Networks"}
{"id": "test_836", "sentence1": "This VLAN could be directing all the traffic with destination address to an Akamai IP address to a dedicated cache server.", "sentence2": "the mobile operators not only can host several MVNO in the same infrastructure but also new business models such as hosting CDN providers.", "label": "reasoning", "domain": "Networks"}
{"id": "test_837", "sentence1": "For the same reason, enhanced CDL system is also expected to use FDMA structure [3]; however typical FDMA structure, which allocates frequency resource to each UAV, cannot support multiple UAVs simultaneously in limited bandwidth.", "sentence2": "enhanced CDL can utilize hybrid FDMA/TDMA structure to improve bandwidth efficiency [5].", "label": "reasoning", "domain": "Networks"}
{"id": "test_838", "sentence1": "Also, each UAV has different size of data traffic.", "sentence2": "if hub performs TDMA scheduling without considering the size of RTT and data traffic, idle resource time and packet delay will occur in hub network.", "label": "reasoning", "domain": "Networks"}
{"id": "test_839", "sentence1": "As shown in (8), sum of RTT and data traffic size in PH should be similar with RTT size in PL.", "sentence2": "in order to perform RTT/traffic based resource scheduling, the condition value Ci, which is shown in (3), should be changed as follows.", "label": "reasoning", "domain": "Networks"}
{"id": "test_840", "sentence1": "In addition, since the number of unused beams is decreased as the number of the UEs increases, the performance of the MAS with dynamic beam control reduces.", "sentence2": "because of the inter-beam interference the performance of the MAS reduces as the number of beams increases.", "label": "entailment", "domain": "Networks"}
{"id": "test_841", "sentence1": "Since antenna tilt angle of switched beamforming is fixed, it is easy to implement and operate.", "sentence2": "the maximum performance of switched beamforming is relatively low than adaptive beamforming.", "label": "contrasting", "domain": "Networks"}
{"id": "test_842", "sentence1": "Since most of the millimeter-wave band is unused, it is possible to construct an ultra-wide band radio access network (RAN) in the 5G mobile communication system.", "sentence2": "in this paper, we have also considered switched beamforming for our proposed scheme.", "label": "neutral", "domain": "Networks"}
{"id": "test_843", "sentence1": "In [6][7], they provide simple closed-form expressions for coverage probability, which is the probability that a randomly deployed UE can achieve a target SINR in heterogeneous network with SBSs.", "sentence2": "this result implies that the number of hidden nodes is proportional to intensity .of SBSs Figure 6: CDF of the number of hidden nodes (without consideration of UE's location) Figure 7: CDF of the number of hidden nodes (with consideration of UE s location) 5.", "label": "neutral", "domain": "Networks"}
{"id": "test_844", "sentence1": "However, the analytical model hardly considers any hidden node problem which is affected by geographical distribution of sender and receiver nodes [9].", "sentence2": "it is hard to use this model to analyze the performance of SBSs with WiFi RAT.", "label": "reasoning", "domain": "Networks"}
{"id": "test_845", "sentence1": "However, results obtained by Eq. (2) does not consider the location of UEs.", "sentence2": "in addition, we also consider the distance distribution of SBSs by using Eq. (4).", "label": "neutral", "domain": "Networks"}
{"id": "test_846", "sentence1": "However, SBSs with multi-RAT have differ­ent characteristics depending on the RAT.", "sentence2": "in order to minimize the risks and failures in real markets, it is es­sential to thoroughly analyze the varying characteristics and performance of multi-RAT SBSs, before their release.", "label": "reasoning", "domain": "Networks"}
{"id": "test_847", "sentence1": "In addition, we only considered the channel for downlink transmission from SBS to UE.", "sentence2": "we will expand our model for estimating the performance of uplink transmission from UE to SBS.", "label": "reasoning", "domain": "Networks"}
{"id": "test_848", "sentence1": "According to AT&T, 3 percent of its smartphone users generate about 40 percent of the network traffc in 2009 [3].", "sentence2": "beside high deploy­ment cost, even after installing additional BSs or upgrading system it may not be very easy for mobile network operators (MNOs) to determine the possible extension in network capacity.", "label": "reasoning", "domain": "Networks"}
{"id": "test_849", "sentence1": "If QcandEF is better than QprevEF, the node regards the EFS message from the previous EF as being dropped from a link error.", "sentence2": "it selects the candidate EF as a new EF if the candidate EF was not a SF in the previous EFS period.", "label": "reasoning", "domain": "Networks"}
{"id": "test_850", "sentence1": "This enables a node that has more SFs to forward an EFS message earlier than those that have fewer SFs.", "sentence2": "a node having more SFs in a previous EFS period has a shorter waiting time than others (the deferred time should sufficiently guarantee the queuing delay and back-off competition of the MAC layer).", "label": "entailment", "domain": "Networks"}
{"id": "test_851", "sentence1": "We assume that if a content requester detects a DATA loss in the CCN module, it retransmits an INTEREST until it receives the DATA successfully.", "sentence2": "as the message forwarding scheme is unreliable the content download time increases due to the INTEREST retransmission.", "label": "reasoning", "domain": "Networks"}
{"id": "test_852", "sentence1": "In addition, the proposed scheme using pseudo-broadcasting shows a slightly higher performance than when using broadcast flooding mode because, in the MAC layer, the retransmission mechanism operates when it fails to transmit so as to provide communication reliability.", "sentence2": "nodes can receive the packet with a higher probability compare to those of broadcast flooding mode.", "label": "reasoning", "domain": "Networks"}
{"id": "test_853", "sentence1": "Although a linear program can be solved in a polynomial time, the caching LP is actually a large-scale problem since there is a large number of contents in the network.", "sentence2": "the caching LP has a special structure, the block-angular form (see Figure 2) which can be decomposed into many subproblems.", "label": "contrasting", "domain": "Networks"}
{"id": "test_854", "sentence1": "The time required to switch the network interface from an active to idle is assumed to be negligible in this work.", "sentence2": "it is sufficient to utilize the time duration where a network interface is turned off as a direct representation of the energy saving for such receiver.", "label": "reasoning", "domain": "Networks"}
{"id": "test_855", "sentence1": "In the 0-1 knapsack problem, we are replacing the variables v with p and w with b to avoid confusion with our defined variables.", "sentence2": "our instance of 0-1 knapsack is defined as following: there are n items xl such that l ∈ [1, n] and xl = 1 if the item is chosen and 0 otherwise.", "label": "reasoning", "domain": "Networks"}
{"id": "test_856", "sentence1": "Comparing with traditional scheduling policies such as: groups with maximum users and firstcome first-served, the gain in service ratio by the proposed algorithm can be at least 2 and 4 times higher, respectively.", "sentence2": "efficient utilization of the expensive and limited wireless spectrum remains an important problem, especially in the context of multimedia streaming services that consume a large portion of the wireless capacity.", "label": "neutral", "domain": "Networks"}
{"id": "test_857", "sentence1": "To offer the flexibility of resource distribution, we are exploiting three type of transmission: unicast, multicast over an SFN, and multicast within the local coverage of a cell.", "sentence2": "our results show that the proposed algorithm can serve up to 40X more users than the common unicast streaming approach.", "label": "neutral", "domain": "Networks"}
{"id": "test_858", "sentence1": "During our implementation, we faced a problem related to the fact that the LTE model in OPNET (which follows the 3GPP spec­ifications for single frequency networks) considers these settings to be static during the entire running time of a simulation.", "sentence2": "the LTE configuration node cannot be adjusted dynamically, and its initial settings (including the MCS assigned for each MBMS bearer) cannot be changed after the deployment phase.", "label": "entailment", "domain": "Networks"}
{"id": "test_859", "sentence1": "Dolphin generates near field assertions via acoustic communications, manipulates the sound power to restrict the communication distance, and uses full-duplex communication [8] to resist relay attacks.", "sentence2": "dolphin has the following prop­erties.", "label": "entailment", "domain": "Networks"}
{"id": "test_860", "sentence1": "For real-time detection tree-based algorithms seem more suitable than SVMs, due to their fast prediction time.", "sentence2": "such good results regarding identification are crucial for future implementations of practical tools.", "label": "neutral", "domain": "Networks"}
{"id": "test_861", "sentence1": "Parking enforcement officers must periodically pa­trol on-street parking areas and check cars one by one to identify cars that are parked over time .", "sentence2": "the design of such a protocol must take into account constrained resources of smart objects and the mobility of drones.", "label": "neutral", "domain": "Networks"}
{"id": "test_862", "sentence1": "After the time period expires, new private keys are generated.", "sentence2": "if a drone is captured, information leakage is limited to the time period during which the private keys were valid.", "label": "reasoning", "domain": "Networks"}
{"id": "test_863", "sentence1": "Securing such communications requires an effective and efficient encryption key establish­ment protocol.", "sentence2": "the design of such a protocol must take into account constrained resources of smart objects and the mobility of drones.", "label": "contrasting", "domain": "Networks"}
{"id": "test_864", "sentence1": "Several pairing-free CL-AKA protocols [10, 9, 26, 24] have thus been proposed.", "sentence2": "most of those protocols were proved to be insecure and only two of them still remain secure: Sun s CL-AKA [24] and Yang s CL-AKA [26].", "label": "contrasting", "domain": "Networks"}
{"id": "test_865", "sentence1": "Since drones and smart objects are battery­powered, energy efficiency as well as security is a critical issue.", "sentence2": "protocol executions should be completed as soon as possible to save energy.", "label": "reasoning", "domain": "Networks"}
{"id": "test_866", "sentence1": "The arms race between malware developers and defenders is endless.", "sentence2": "it is essential to continuously track and understand the latest strategies of attackers in manipulating botnets for attacks.", "label": "reasoning", "domain": "Networks"}
{"id": "test_867", "sentence1": "Probabilistic forwarding (e.g., Crowds [?]) delinks messages from their actual senders without a preset forwarding path.", "sentence2": "the onus of message confidentiality in Crowds is on the probabilistic forwarding nodes since those nodes have message decryption keys.", "label": "contrasting", "domain": "Networks"}
{"id": "test_868", "sentence1": "In this protocol, a constrained yet PUF-enabled device and a server compute the session key and authenti­cate each other.", "sentence2": "the device authenticates itself by performing a zero knowledge proof on its PUFs generated secrets.", "label": "entailment", "domain": "Networks"}
{"id": "test_869", "sentence1": "The priorities are considered in regards to both the sites corresponding to each distributed active participant, and also the cameras within each site.", "sentence2": "hosseini et al [4, 3] adopted priority-based approaches towards the study of e.ciently transmitting, rendering, and displaying bulky 3D gaming information to power-limited devices given the importance of di.erent 3D objects in the gaming context.", "label": "neutral", "domain": "Networks"}
{"id": "test_870", "sentence1": "As can be seen, the larger value of Rmax results in more bandwidth reduction, and thus quality sacrifices, for streams in C22 (corresponding to the highest priority class), while preserving more of the full-bandwidth streams in C11 (corresponding to the least priority class).", "sentence2": "our framework does provide quality degradation in general, but considering the bandwidth savings achieved using our adaptations, it is reasonable to believe that teleimmersive users would make this sacrifice in quality in ex­change for respecting their view and the network bandwidth constraints.", "label": "neutral", "domain": "Networks"}
{"id": "test_871", "sentence1": "The experiments used a Logitech C920 HD webcam [11] mounted on an Ascending Technologies (AscTec) Firefly [3] UAV (shown in Figure 1).", "sentence2": "802.11 was never intended to be used for fast moving and direction changing devices.", "label": "neutral", "domain": "Networks"}
{"id": "test_872", "sentence1": "He presents a proxy server that can adapt SVC video streams according to current network conditions.", "sentence2": "“instead of reacting on packet loss, [the] approach uses an increase in queueing delay at the router to detect phases of throughput degradation.” [9] ", "label": "contrasting", "domain": "Networks"}
{"id": "test_873", "sentence1": "An alternative approach is to recreate RDMA state and resources on the newly mi­grated machine [31].", "sentence2": "implementation complex­ity of such approaches prohibits their deployment in a dis­tributed, general-purpose virtualized infrastructure.", "label": "contrasting", "domain": "Networks"}
{"id": "test_874", "sentence1": "However, in contrast to HyV, vRDMA uses a paravirtual communication channel to perform data operations.", "sentence2": "vRDMA cannot completely bypass the hypervisor on data operations.", "label": "entailment", "domain": "Networks"}
{"id": "test_875", "sentence1": "HyV allows for a more efficient memory usage because all the RDMA memory (which is DMA able) is explicitly regis­tered through the hypervisor.", "sentence2": "hyV has a much lower memory footprint than passthrough, and thereby al­lows packing more VMs on to a single physical machine.", "label": "reasoning", "domain": "Networks"}
{"id": "test_876", "sentence1": "In the pa­per, we provide a detailed evaluation of HyV for different RDMA technologies and operations.", "sentence2": "the critical aspect of RNIC s is that the control paths are used to directly map device provided, connection­specific, structures into an applications address space.", "label": "neutral", "domain": "Networks"}
{"id": "test_877", "sentence1": "Here, the provider also expects a contiguous (host) virtual memory region, but unlike in bottom up it does not back the region itself but relies on the operating systems to do so.", "sentence2": "the provider only pins the memory to ensure that the pages can be used for DMA transfers and extracts them for installation on the device.", "label": "reasoning", "domain": "Networks"}
{"id": "test_878", "sentence1": "In this work, we presented HyV, a virtualization framework that unleashes the full performance advantages of RDMA interconnects to virtual machines.", "sentence2": "hyV is implemented as a plugin component for the Linux OFED RDMA software stack.", "label": "neutral", "domain": "Networks"}
{"id": "test_879", "sentence1": "This strategy has the downsides that, when a video streaming is aborted (for instance to start streaming a new video), it is often the case that the hottest HTTP/1.1 connection is aborted to stop receiving the next video segment, in which case the next video streaming will start with a cold TCP connection and will not benefit of any prior warm up. ", "sentence2": "with HTTP/2, this case no longer happens and the streaming of a first video will benefit to the next one.", "label": "contrasting", "domain": "Networks"}
{"id": "test_880", "sentence1": "To determine the origin of this performance gain of HTTP/2, a deeper analysis is necessary.", "sentence2": "when using the traditional adaptation (no push and bandwidth estimation is based on downloading time), the client directly selects the representation with the highest bitrate and continues using it.", "label": "neutral", "domain": "Networks"}
{"id": "test_881", "sentence1": "HTTP/2 defines a general multiplexed transport designed mainly for HTTP messages, but that can accommodate other messaging application protocols.", "sentence2": "we can use HTTP/2 framing layer to transport WebSocket-like messages over an HTTP/2 stream rather than over a dedicated TCP connection.", "label": "reasoning", "domain": "Networks"}
{"id": "test_882", "sentence1": "Admittedly, Halma requires multiple antenna elements at the transmit­ter side, entailing more space cost.", "sentence2": "we have observed vast energy saving from Halma even with 2 transmit antenna elements, with marginal space cost.", "label": "contrasting", "domain": "Networks"}
{"id": "test_883", "sentence1": "The receiver is randomly placed within line-of-sight of the transmitter (which may adversely increase channel similarity) in an office environment.", "sentence2": "when both channel magnitude and phase are used as signatures, on average in 95.8% of cases, an antenna's instantaneous signature remains a best-match with its other signatures.", "label": "neutral", "domain": "Networks"}
{"id": "test_884", "sentence1": "In some sense, Halma actually augments amplitude modulation on legacy ZigBee by lever­aging the symbol amplitude variation naturally provided by negligible compared with the switching period in Halma (8 samples or 4 µs for ZigBee).", "sentence2": "halma's fine­grained, sub-symbol-level antenna switching mechanism is feasible in practice.", "label": "reasoning", "domain": "Networks"}
{"id": "test_885", "sentence1": "Since ZigBee uses a differential demodulator to decode normal data symbols, and antenna switching occurs only per Ns = 8 samples, AIC itself is unlikely to affect the performance of the normal decoder.", "sentence2": "the receiver decodes the normal data symbols first, separately from antenna index decoding.", "label": "reasoning", "domain": "Networks"}
{"id": "test_886", "sentence1": "The experiments are conducted in an busy office environment with 12 people, and 2 intentionally walking back and forth.", "sentence2": "different transmit antennas symbols are dis­torted by the channel in different ways.", "label": "neutral", "domain": "Networks"}
{"id": "test_887", "sentence1": "(iii) The optimal configuration may vary due to channel variation.", "sentence2": "the receiver monitors the throughput TH(t) for current configuration.", "label": "reasoning", "domain": "Networks"}
{"id": "test_888", "sentence1": "We observe that the modulated waveform of any chip se­quence is made from 4 elementary patterns, corresponding to 32 complex samples (Figure 6 shows two of such patterns).", "sentence2": "halma employs an adap­tive antenna hopping (AAH) protocol that efficiently selects the subset of antennas to optimize this tradeo., based on a model-driven framework instrumented by channel pro.le measurement.", "label": "neutral", "domain": "Networks"}
{"id": "test_889", "sentence1": "In contrast, Halma focuses on achieving high energy efficiency.", "sentence2": " for a ZigBee link with single RF-chain transmitter and receiver, Halma can scale link capacity with Nt at a even faster rate than SSK, which translates into enormous energy saving.", "label": "entailment", "domain": "Networks"}
{"id": "test_890", "sentence1": "The USB adapter's TX, RX and idle power consumption all grows linearly as the number of active antennas increases.", "sentence2": "with 3 antennas, the idle power is 1.3× that of 1-antenna case.", "label": "entailment", "domain": "Networks"}
{"id": "test_891", "sentence1": "We placed 4 DEE R loggers, complete with modified collar, on free­ranging horses in a fenced area of maximum length 150 m. This allowed us to test the effect of the animal body on contact detec­tion.", "sentence2": "since animals were free to range, we could test the effect of neither distance from ground nor relative horse body po­sitions on contact detection.", "label": "contrasting", "domain": "Networks"}
{"id": "test_892", "sentence1": "The camera failed again on April 14, then we removed it.", "sentence2": "our camera trap dataset covers ~15 days over two separate periods.", "label": "reasoning", "domain": "Networks"}
{"id": "test_893", "sentence1": "The data recorded by nodes are timestamped with the logical time (epoch) of acquisition.", "sentence2": "the logical time is synchronized neither across nodes nor w.r.t. global (physical) time.", "label": "contrasting", "domain": "Networks"}
{"id": "test_894", "sentence1": "In this case, the contact duration affects only the radio contribution, explaining the gentle slope for Tc > 60'.", "sentence2": "the last component of the WILDSCOPE toolset is the database where the data gathered is stored.", "label": "neutral", "domain": "Networks"}
{"id": "test_895", "sentence1": "The link between PCO-based synchronization methods and consensus algorithms for networked oscillators is established in [32].", "sentence2": "only limited work has been focused on extending distributed desynchronization to the multichannel case [7].", "label": "contrasting", "domain": "Networks"}
{"id": "test_896", "sentence1": "If no ticket key has been configured, a fresh one is generated for each IP address and port (but not for each virtual host).", "sentence2": "if a ticket key is specified in the global configuration of the server, all tickets created by any virtual host can be resumed on any other.", "label": "contrasting", "domain": "Networks"}
{"id": "test_897", "sentence1": "However, non-public files cannot be served from this low-trust origin when the user wants to download data from her account, because they require access to the session cookie to prove that the user is authorized to view the file.", "sentence2": "the dl-web.dropbox.com origin is used for the purpose of displaying files from the user's own Dropbox account while he is logged in.", "label": "reasoning", "domain": "Networks"}
{"id": "test_898", "sentence1": "On the server, each parameter is considered separately in a manually configured set of complex rules to determine the virtual host that will handle the request (see Section 3 for more detail).", "sentence2": "most web servers will pick a fallback virtual host when the normal routing rules fail.", "label": "entailment", "domain": "Networks"}
{"id": "test_899", "sentence1": "Many web servers em­ploy virtual hosting to serve multiple HTTPS domains be­hind the same TLS server.", "sentence2": "the same check applies to every other pair of IP address and port where this certificate is used.", "label": "neutral", "domain": "Networks"}
{"id": "test_900", "sentence1": "Figure 2 depicts how HTTPS requests are processed by Akamai: each PoP has N custom certificates installed for N virtual hosts, and each certificate is served on a dedicated IP address. ", "sentence2": "if a client connects to IP 1, it will be given the certificate for a.com, whereas if it connects to IP 2, it will be given the certificate for *.z.com.", "label": "reasoning", "domain": "Networks"}
{"id": "test_901", "sentence1": "We have described four different exploits of virtual host confusion against major websites.", "sentence2": "these particular exploits do not give a clear picture of the general proportion of all websites vulnerable to similar attacks.", "label": "contrasting", "domain": "Networks"}
{"id": "test_902", "sentence1": "In each subfigure, x-axis and y-axis vary the desirable data demand u and value v of the user types.", "sentence2": "each point in the subfigures represents a unique user type.", "label": "entailment", "domain": "Networks"}
{"id": "test_903", "sentence1": "We observe that although the resulting data load under any data cap gI does not vary much, the maximum revenue R * I (gI ) varies significantly.", "sentence2": "when the pricing converges to the flat-rate structure as gI goes to 1, the optimal revenue decreases to a minimum value.", "label": "entailment", "domain": "Networks"}
{"id": "test_904", "sentence1": "As we observe from the plot, when apps react to firewall-timer, and when firewall-timer is set to a high value, it substantially reduces signaling per minute for most of the apps.", "sentence2": "for apps having long-lived background connections such as Skype and Facebook, a high value of firewall timeout reduces the signaling overhead by more than a factor of 4.", "label": "entailment", "domain": "Networks"}
{"id": "test_905", "sentence1": "A desirable way to collect traffic data of apps is to access their packet traces in operator's network.", "sentence2": "we do not have access to such data as it is typically not published by network operators.", "label": "contrasting", "domain": "Networks"}
{"id": "test_906", "sentence1": "As a increases from 2.4 to 3.6, most of the links become local, short-range links and the network approaches the regular multi-hop pattern that also results in high EDP.", "sentence2": "we need to choose a value of a that achieves a compromise between these two extreme cases.", "label": "reasoning", "domain": "Networks"}
{"id": "test_907", "sentence1": "In this way we are able to bring physically far and highly interacting nodes to logical proximity so that they are easily accessible to each other.", "sentence2": "in this context, design of small-world network-based NoC architectures [1] is a notable example.", "label": "neutral", "domain": "Networks"}
{"id": "test_908", "sentence1": "To reduce complexity, we restrict the routing algorithm to X-Y routing (dimension-order routing) with the standard 2D mesh.", "sentence2": "the path of a packet has at most one turn.", "label": "reasoning", "domain": "Networks"}
{"id": "test_909", "sentence1": "Through the website, users can define emergent monitoring tasks and preview the video results immediately.", "sentence2": "the website consists of three units: setting, execution, and display.", "label": "entailment", "domain": "Networks"}
{"id": "test_910", "sentence1": "Real-time video surveillance systems are essential in modern 2D environments has been studied in [1, 2].", "sentence2": "in this prototype, the model of PTZ cameras is Compro IP540/IP570[6], which supports the maximum resolution of 1280 by 1024 and provides flexible configurations (e.g., 340-degree pan, 100-degree tilt, and 12x optical zoom).", "label": "neutral", "domain": "Networks"}
{"id": "test_911", "sentence1": "For example, 480p video with a 16:9 display aspect ratio can not have a square pixel shape since (480/9) × 16 is not an integer.", "sentence2": "it consumes even more power than 540p.", "label": "reasoning", "domain": "Networks"}
{"id": "test_912", "sentence1": "Video pass-through, combined with batching/prefetching, can save 52% to 61% for continuous video streaming.", "sentence2": "contention does vary the relative contribution of different optimization mechanisms.", "label": "neutral", "domain": "Networks"}
{"id": "test_913", "sentence1": "U-Wear offers a training-based channel estimation approach that requires the presence of a training sequence known a-priori in the transmitted packet.", "sentence2": "u-Wear leverages the good autocorrelation property of the synchronization preamble sequence, discussed in Section 3.1.2, to estimate the CIR.", "label": "entailment", "domain": "Networks"}
{"id": "test_914", "sentence1": "Finally, the dielectric nature of the human body also affects the coupling between on-body RF antennas and the body itself.", "sentence2": "the gain and the radiation pattern of the antenna deteriorate because of the contact or proximity with the human body [16], while the resonant frequency and the input impedance of the antenna may shift from their nominal values.", "label": "entailment", "domain": "Networks"}
{"id": "test_915", "sentence1": "The ADMP401 offers a mostly flat bandwidth, i.e., -3 dB roll o., between 100 Hz and 15 kHz, omnidirectional sensitivity pattern, and requires a supply voltage between 1.5 V and 3.3 V DC.", "sentence2": "as a proof of concept, we design two prototypes that im­plement the U-Wear framework and operate in the near­ultrasonic frequency range, i.e., 17 - 22 kHz, using COTS speakers and microphones.", "label": "neutral", "domain": "Networks"}
{"id": "test_916", "sentence1": "However, it is not intuitive to users how much they have to rotate their wrists in order to move the pointers in an intended direction, which makes them fail to reach the target.", "sentence2": "users have trouble moving a pointer in a diagonal direction, as it requires them to rotate their wrist in two axes simultaneously.", "label": "entailment", "domain": "Networks"}
{"id": "test_917", "sentence1": "One of the challenges in RF phase based tracking is accurate measurement of the received signal phase.", "sentence2": "the carrier frequency offset (CFO) between the sender and the receiver causes the phase to change over time even if the receiver is not moving.", "label": "entailment", "domain": "Networks"}
{"id": "test_918", "sentence1": "We believe this approach is reasonable in normal situations since such idle connections will be likely to stay paused to minimize the cellular traffic usage.", "sentence2": "we understand this is not a solution to a Denial-of-Service (DoS) attack on D2Prox, which would require more careful thoughts.", "label": "contrasting", "domain": "Networks"}
{"id": "test_919", "sentence1": "The disparity comes from the fact that the develop­ers have to handle each case differently without the support from the transport layer.", "sentence2": "vLC exploits D2BufMgr in D2TP by setting an appropriate deadline, and delays cellular data transfer if possible to offload to Wi-Fi.", "label": "neutral", "domain": "Networks"}
{"id": "test_920", "sentence1": "Higher-resolution cameras capture more pix­els on the transmitter screen, and thus they support smaller grids on the transmitter screen and achieve higher throughput.", "sentence2": "these observations together justify the configuration in HiLight (Algorithm 1), which increases the for dark areas and dynamic scenes.", "label": "neutral", "domain": "Networks"}
{"id": "test_921", "sentence1": "To address the challenge, we design strategies to filter out color intensity change associated with interfering factors, and adapt our strategy to the current scene type.", "sentence2": "for the static scene, we leverage an audio beamforming algorithm [33, 35] to minimize the impact of interfering factors and extract the desired color inten­sity change encoded with data; for the dynamic scene, we identify patterns caused by the encoded change.", "label": "entailment", "domain": "Networks"}
{"id": "test_922", "sentence1": "Higher-resolution cameras capture more pix­els on the transmitter screen, and thus they support smaller grids on the transmitter screen and achieve higher throughput.", "sentence2": "the high-end SLR camera supports 720 grids on the 10-in screen enabling 6.6 Kbps throughput, six times higher than that achieved by iPhone 5s in our prototype.", "label": "entailment", "domain": "Networks"}
{"id": "test_923", "sentence1": "Among different screen types, OLED screens are the most preferable, because OLED screens do not have backlight and each pixel emits light independently.", "sentence2": "colors on OLED screens are brighter with higher contrast, making color intensity change easier to detect.", "label": "reasoning", "domain": "Networks"}
{"id": "test_924", "sentence1": "We obtain the noise correlation matrix R via a training process for a static scene.", "sentence2": "once the receiver detects a static scene, it buffers frames for M frame windows.", "label": "entailment", "domain": "Networks"}
{"id": "test_925", "sentence1": "We encode data by changing the pixel transparency of the communication layer.", "sentence2": "this is because these steps are performed upon sampled pixels of each frame.", "label": "neutral", "domain": "Networks"}
{"id": "test_926", "sentence1": "A three-layer FatTree with K-port switches can provide up to (K/2)2 shortest paths between every pair of endpoints that locate in different pods, e.g., a FatTree with 24-port switches, contains 144 shortest paths between a pair of endpoints, hence MPTCP leverages the redundant paths in the topology by opening multiple subflows to achieve the goal of high bandwidth utilization.", "sentence2": "opening more subflows is expensive due to the following three ways: (1) spreading data across multiple subflows leads to more resource and scheduling overheads.", "label": "contrasting", "domain": "Networks"}
{"id": "test_927", "sentence1": "The increase of CPU usage is mainly due to the increase in the number of the subflows, because it need more CPU cycles to schedule the transmissions of all the subflows.", "sentence2": "it does not distinguish mice flows and elephant flows, and thus unnecessarily opens too many subflows for mice flows. ", "label": "neutral", "domain": "Networks"}
{"id": "test_928", "sentence1": "The trace data were then divided into time intervals with a fixed length of 100 s. To quantify the effectiveness of using past bandwidth statistics to predict future values, we measured the bandwidth similarity of two vectors with PBA.", "sentence2": "we chose a set of intervals, the candidate prediction intervals, which ranged from 50 ms to 2.6 s in length and spaced 100 ms apart.", "label": "entailment", "domain": "Networks"}
{"id": "test_929", "sentence1": "To generalize the model, we assume that the update interval is equal to the RTT, but we can derive similar results if the update interval is equal to the interval-arrival time of the ACKs or a fixed interval.", "sentence2": "we set up the network topology shown in Fig. 1 with a link buffer size of 852 packets [20] and a base propagation delay of 160 ms (Dmin=0.16 s) as we measured from an actual 3G/HSPA network.", "label": "neutral", "domain": "Networks"}
{"id": "test_930", "sentence1": "The QCC queue length adaptation algorithms can be divided into two key stages: 1) the startup phase and 2) the adaptive phase.", "sentence2": "the startup phases of TCP Vegas and FAST TCP are identical to the TCP slow-start phase.", "label": "entailment", "domain": "Networks"}
{"id": "test_931", "sentence1": "Not much work has been done on determining stable paths or trees that could exist for a longer time.", "sentence2": "to the best of our knowledge, there has been no work done to determine stable sequence of a communication topology that spans (i.e., connect all the SU nodes) the entire CRAHN network of SU nodes.", "label": "entailment", "domain": "Networks"}
{"id": "test_932", "sentence1": "In general, for k such parallel routes we would have, the probability of breakage as Pk = (P1)k where P1 is probability of breakage of single route.", "sentence2": "the probability of communication of breakage between Source and Destination is reduced exponentially if parallel routes are used.", "label": "reasoning", "domain": "Networks"}
{"id": "test_933", "sentence1": "Namely, the solution concept for this market is related to the competitive (or, Walrasian) equilibrium [3], [18], which has been also applied in communication networks [4], and extended to graphical economies (which exhibit localities ) [16], [17].", "sentence2": "for the problem under consideration, there do not exist explicit price variables (or, price signals), and hence we employ a different equilibrium concept: Definition 5.", "label": "contrasting", "domain": "Networks"}
{"id": "test_934", "sentence1": "Therefore, the lex-optimal allocation is an exchange equilibrium, and, additionally, any possible exchange equilibrium is also a lex-optimal allocation.", "sentence2": " the competitive interactions of rational users embedded in a graph, lead to the same allocation point that a central designer would have selected for such a system.", "label": "entailment", "domain": "Networks"}
{"id": "test_935", "sentence1": "This important result reveals that there is a unifying approach that solves the resource allocation problem for graph-constrained systems (or, economies), for different node behaviors.", "sentence2": "we can apply the max-min fair criterion, that has been extensively used for load balancing in centralized communication networks (e.g., see [21] and references therein), to service exchange models with autonomous and selfish nodes.", "label": "entailment", "domain": "Networks"}
{"id": "test_936", "sentence1": "A mixed-signal SIC architecture has been proposed in [28], where the digital TX signal is processed and upconverted to RF for cancellation.", "sentence2": "this requires a separate upconversion path which introduces its own noise and distortion. ", "label": "contrasting", "domain": "Networks"}
{"id": "test_937", "sentence1": "This work explores logically centralized inter-domain QoS routing mediators for the provisioning of inter-domain path guarantees in light of ongoing changes in the Internet ecosystem.", "sentence2": "copyrights for third-party components of this work must be honored.", "label": "neutral", "domain": "Networks"}
{"id": "test_938", "sentence1": "We thus propose a sample-select approach to heuristically tackle the QMRP in an online manner.", "sentence2": "given the NP-hardness of the optimal path calculations, we employ a sample-select pro­cess, where in the first stage, a set of feasible paths is sampled (i.e., generated) in polynomial time, and subsequently one of them is selected for the actual embedding.", "label": "entailment", "domain": "Networks"}
{"id": "test_939", "sentence1": "This is because content providers utilize content distribution networks for serving content at different user locations.", "sentence2": "a problematic content server can only affect the performance of a subset of user locations.", "label": "reasoning", "domain": "Networks"}
{"id": "test_940", "sentence1": "This is because content providers utilize content distribution networks for serving content at different user locations.", "sentence2": "we use robust regression for this baseline modeling because it can minimize the impact of extreme outlier data points on the produced model.", "label": "neutral", "domain": "Networks"}
{"id": "test_941", "sentence1": "More and more devices in the future, such as Google Glasses, baby monitors, and new generation of home appliances, will all desire mouse functionalities, which allow users to choose from a wide variety of options and easily click on different parts of the view.", "sentence2": "a traditional mouse, which requires a flat and smooth surface to operate, cannot satisfy many new usage scenarios.", "label": "contrasting", "domain": "Networks"}
{"id": "test_942", "sentence1": "The ITS is a critical infrastructure, and any benign or ma­licious fault could have far-reaching consequences.", "sentence2": "reliability, security and survivability are of paramount im­portance.", "label": "reasoning", "domain": "Networks"}
{"id": "test_943", "sentence1": "This type of jammer simply emits radio signals continuously (e.g. random noise), which interferes with the signal, i.e., it decreases the signal-to-noise ratio.", "sentence2": "one of the main objectives of these technologies is to increase safety.", "label": "neutral", "domain": "Networks"}
{"id": "test_944", "sentence1": "We use the example of remote control attack to illustrate the game between the system owner and the attacker.", "sentence2": "mCUs cannot directly interpret the encrypted data, on-the-fly decryption extracts the program from the encrypted code.", "label": "neutral", "domain": "Networks"}
{"id": "test_945", "sentence1": "The larger the window, the more likely they will choose different contention time so as to avoid collisions.", "sentence2": "this will not work for prioritized contention, since the contention time is not randomly picked but pre-defined by the alignment metrics.", "label": "contrasting", "domain": "Networks"}
{"id": "test_946", "sentence1": "On the other hand, this prioritized contention causes an extra difficulty in avoiding collisions.", "sentence2": "when different users have similar alignment metrics, they will end up in collision since they have similar contention time.", "label": "entailment", "domain": "Networks"}
{"id": "test_947", "sentence1": "After user selection, the AP needs to know the exact CSI to perform precoding before beamforming to users.", "sentence2": "rBF only precodes with the randomly selected directions without CSI feedback, which causes much capacity loss.", "label": "contrasting", "domain": "Networks"}
{"id": "test_948", "sentence1": "Fig. 11 presents the average throughput under 3 scenarios.", "sentence2": "they should coordinate with each other spontaneously to form the best user group for each UL MU-MIMO transmission.", "label": "neutral", "domain": "Networks"}
{"id": "test_949", "sentence1": "For this case, later users cannot detect that more than one users CAs collide on the same subcarrier, and therefore will not start collision recovery.", "sentence2": "these two kinds of collisions could be limited to a minimum using a reasonable contention window.", "label": "contrasting", "domain": "Networks"}
{"id": "test_950", "sentence1": "Due to frequency diversity, the subcarriers may have different channel gains, which makes the user selection problem more sophisticated.", "sentence2": "for each user, one subcarrier s1 may be better aligned with a Signpost direction, whereas another subcarrier s2 may be better aligned to a different direction.", "label": "entailment", "domain": "Networks"}
{"id": "test_951", "sentence1": "This fact indicates that having exact CSI is critical for DL MU-MIMO beamforming in practice.", "sentence2": "recently, many experimental studies [9–18] have explored MU-MIMO in practical scenarios.", "label": "neutral", "domain": "Networks"}
{"id": "test_952", "sentence1": "Since U3 and U4 are correlated to AP1’s selected users (U1 and U2), they may reduce total network capacity when selected by AP2. ", "sentence2": "aP2 excludes U3 and U4, and instead selects a bridge client (U6) uncorrelated to all currently selected users w.r.t.", "label": "reasoning", "domain": "Networks"}
{"id": "test_953", "sentence1": "Their design principle is to make the most of the multiple DoFs at the client side.", "sentence2": " Kardia focus on the optimal utilization of the DoFs at the AP by employing multi-cell transmission to combat the channel hardening effect.", "label": "contrasting", "domain": "Networks"}
{"id": "test_954", "sentence1": "Multi-user MIMO (MU-MIMO) is a hallmark of the recent WLAN standard 802.11ac, with which an Mtx-antenna AP can simultaneously transmit up to Mtx data streams to different users, thus achieving a maximum degree-of-freedom (DoF) of Mtx.", "sentence2": "exploiting MU-MIMO's full potential is a non-trivial issue in practice.", "label": "contrasting", "domain": "Networks"}
{"id": "test_955", "sentence1": "The winner's CSI is in turn leveraged to construct the next probing packet.", "sentence2": "hence, the degree of channel orthogonality between a pair of users will also vary across subcarriers.", "label": "neutral", "domain": "Networks"}
{"id": "test_956", "sentence1": "However, when other nodes are added to it as relays, cycles might be formed.", "sentence2": "when paths connecting different pairs of multicast members share the same relay nodes, such node may receive redundant information, which indicates that cycles come into being.", "label": "entailment", "domain": "Networks"}
{"id": "test_957", "sentence1": "When modified neighbor selection criteria is applied, more constraints are added to neighbor selection.", "sentence2": "the distance between a node and its selected neighbor may in­crease.", "label": "reasoning", "domain": "Networks"}
{"id": "test_958", "sentence1": "The spectrum sensing approach, which is less widely used, relies on the user devices to perform spectrum sens­ing.", "sentence2": "it requires the devices to be equipped with proper sensing hardware and to have enough power for sensing and signal analysis.", "label": "reasoning", "domain": "Networks"}
{"id": "test_959", "sentence1": "It is indeed a creative way to utilize the similarity relationship between locations.", "sentence2": "in our mechanism, we consider not only the similarity relationship but also the linear dependence between locations.", "label": "contrasting", "domain": "Networks"}
{"id": "test_960", "sentence1": "Instead of explicitly aligning Pilotfish symbols to incumbent symbols, we prefer truely independent transmissions.", "sentence2": "in our design we set Pilotfish's symbol size to be twice of the incumbent symbol, as shown in Figure 6(a).", "label": "reasoning", "domain": "Networks"}
{"id": "test_961", "sentence1": "We expect these new channels will bring significant value, because communication spectrum is quite expensive today.", "sentence2": "at the receiving side, if we pick the FFT outputs (i.e. take samples) corresponding to these elements, the received OFDM symbol will be entirely nulled", "label": "neutral", "domain": "Networks"}
{"id": "test_962", "sentence1": "The downside of such an IIR filter, however, is that the signal will spread out in the time domain.", "sentence2": "it is not an ISI-free filter.", "label": "reasoning", "domain": "Networks"}
{"id": "test_963", "sentence1": "Only when the SNR approachs to marginal value, e.g 5dB, the false negtive rate increases a little to 20%.", "sentence2": "we conclude that the CCA for Pilotfish MAC is robust.", "label": "reasoning", "domain": "Networks"}
{"id": "test_964", "sentence1": "The goal of our filter design is that the signal leakage at the stop-band would be as low as noise floor (i.e., -90dBm), and thus the interference to nearby data communication can be neglected.", "sentence2": "we recorded received signal strength indicator (RSSI) on every link, then calculated SINR in guardband using a model with three nodes: one node as guardband transmitter, one as guardband receiver and one as incumbent transmitter.", "label": "neutral", "domain": "Networks"}
{"id": "test_965", "sentence1": "In a typical real-world scenario one may have tens or hun­dreds of femto-BSs inside a macrocell and hundreds or thou­sands of users.", "sentence2": "femto-BSs would be grouped into clusters of nearby femto-BSs which can concurrently serve a number of users.", "label": "reasoning", "domain": "Networks"}
{"id": "test_966", "sentence1": "Let us consider our graph example.", "sentence2": "it can verify if a packet has already been received.", "label": "neutral", "domain": "Networks"}
{"id": "test_967", "sentence1": "An­other impact is the high number of collisions between the nodes which try to transmit.", "sentence2": "this high load impacts not only the number of received packets within deadline, but also the energy consumption and the end-to­end delay as shown in all other figures.", "label": "reasoning", "domain": "Networks"}
{"id": "test_968", "sentence1": "On the other hand, an increasing number of applications re­quire additional guarantees, like minimal delay and reliabil­ity, for the packets they generate.", "sentence2": "a good routing protocol needs to select best paths through several exist­ing ones to balance energy and memory consumption and guarantee the application s constraints.", "label": "reasoning", "domain": "Networks"}
{"id": "test_969", "sentence1": "Moreover, few of them have investigated the method to estimate the quality of sensing data.", "sentence2": "we systemically consider the participants’ willingness to take a sufficient amount of efforts in crowdsensing, and bridge the gap between quality of sensing data and rewards for contributions, by providing a quality based incentive mechanism.", "label": "contrasting", "domain": "Networks"}
{"id": "test_970", "sentence1": "This mechanism estimates the quality of sensing data, and offers each participant a reward based on her effective contribution.", "sentence2": "lee and Hoh [13] proposed a reverse auction based dynamic pricing scheme to motivate participants to sell their sensing data with claimed bids.", "label": "neutral", "domain": "Networks"}
{"id": "test_971", "sentence1": "Sensing, processing, and transmitting data in crowdsensing applications requires manual efforts and physical resources.", "sentence2": "appropriate rewards are always expected to compensate the owners of task-taking mobile devices.", "label": "reasoning", "domain": "Networks"}
{"id": "test_972", "sentence1": "In addition, it shows little consideration about incentives.", "sentence2": "although various empirical experiments [17, 20, 25, 35] demonstrate that financial and social incentives do have an impact on the performance of participants, such as engagement, compliance and quality, they fail to generalize an incentive model to adaptively guide the participants’ behavior.", "label": "contrasting", "domain": "Networks"}
{"id": "test_973", "sentence1": "We consider goodput over throughput as different protocols incur different amounts of protocol overhead, and goodput can better reflect the data transmission efficiency and delay, i.e., a higher goodput indicates better efficiency and a shorter delay [33].", "sentence2": "to understand the protocol overhead, we further evaluate the detailed breakdown of throughput for each protocol in this section.", "label": "contrasting", "domain": "Networks"}
{"id": "test_974", "sentence1": "Recall that the transmitter in each D2D link randomly selects a channel from the set {m1,m2,...,m|M|} for data transmission, while one-hop cellular users transmit in chan­nels {n1,n2,...,n|N|}.", "sentence2": "there is no cross-tier in­terference between cellular links and D2D links.", "label": "reasoning", "domain": "Networks"}
{"id": "test_975", "sentence1": "So far, D2D communication has been mostly studied for direct downlink traffic offloading between two mobile users within proximity, such as interference management, resource allocation, performance evaluation, etc.", "sentence2": "for a cellular user, the requested data is delivered from the buffer of a neighboring user via D2D communication, without traversing through the serving BS or the core network.", "label": "entailment", "domain": "Networks"}
{"id": "test_976", "sentence1": "Note that in previous works [4, 7, 8], to facilitate the analysis of coverage probability and average rate for a user occupying a given channel, it assumed a large enough user density such that the channel will always be occupied in all cells, while neglecting the fact that it actually becomes more difficult for a user to obtain a channel as the user density increases.", "sentence2": " the users may suffer from “channel outage” due to the limited frequency resources (channels) in a cell. ", "label": "entailment", "domain": "Networks"}
{"id": "test_977", "sentence1": "Social networks are the most engaging applications on mobile devices, and they are becoming the main sources for users to consume content.", "sentence2": "content retrieval, especially for embedded links and multimedia, can often be too slow, too energy hungry or too expensive for on-the-go mobile users.", "label": "contrasting", "domain": "Networks"}
{"id": "test_978", "sentence1": "Also, notice that since WiFi is often not available when our users launch Twitter, prefetch at application launch can be costly in terms of data and battery usage.", "sentence2": "simply prefetch at launch does not work as well.", "label": "reasoning", "domain": "Networks"}
{"id": "test_979", "sentence1": "Because existing work has carefully studied the network-condition-based prefetching scheduling using fine-grained traces [27, 21], we focus our work and the corresponding data collection on the complementary com­ponents.", "sentence2": "to limit the collection overhead and to avoid user burn out, we decide not to collect detailed net­work trace in a fine-grain manner, but only log the WiFi availability, which is a major leverage for prefetching.", "label": "reasoning", "domain": "Networks"}
{"id": "test_980", "sentence1": "We achieve this goal by selecting an appropriate threshold and time-window.", "sentence2": "only links with a click-through score higher than the threshold (based on bud­get §Tuning the Thresholds) will be prefetched.", "label": "entailment", "domain": "Networks"}
{"id": "test_981", "sentence1": "Although Twitter s tweets are publicly avail­able, when and how users access Twitter is not.", "sentence2": "we collected a large set of usage data from Twidere1 users who agreed to provide this information to us.", "label": "reasoning", "domain": "Networks"}
{"id": "test_982", "sentence1": "We select as many as possible features in the initial set to be comprehensive.", "sentence2": "for each individual user, features without good distinguish power introduce noise in the prediction.", "label": "contrasting", "domain": "Networks"}
{"id": "test_983", "sentence1": "The increasing popularity and ubiquity of such content consumption calls for exceptional support from mobile devices.", "sentence2": "the current experience of content consumption is far from satisfactory in terms of access delay and network availability.", "label": "contrasting", "domain": "Networks"}
{"id": "test_984", "sentence1": "In particular, we compute average bits/sec/H z/m2 for each UHF channel and present average values of the 20 channels normalized to the achievable rate of a UHF channel that has no TV transmit­ters nor TV receivers.", "sentence2": "the channel switching time is defined as the duration between when the channel switching command is sent and when the new TV programming is displayed.", "label": "neutral", "domain": "Networks"}
{"id": "test_985", "sentence1": "Unfortunately, this technique alone would provide little benefit to WATCH due to the strong interfer­ence from TV transmitters.", "sentence2": "we design CAT, a Constructive Addition Transmission scheme that maxi­mizes secondary signal SINR at SU-RX after accounting for the channels from the TV transmitter to SU-RX.", "label": "reasoning", "domain": "Networks"}
{"id": "test_986", "sentence1": "Instead, we estimate , which HP U 2 does not need to use preambles of TV signals and therefore does not require x between the primary and the secondary system.", "sentence2": "the SU network operates fully asynchronously.", "label": "entailment", "domain": "Networks"}
{"id": "test_987", "sentence1": "However, channel changes of TV receivers cannot be detected by external techniques such as spectrum sensing.", "sentence2": "we require active TV receivers to inform the database controller of the channel changes through primary feedback.", "label": "reasoning", "domain": "Networks"}
{"id": "test_988", "sentence1": "To date, TVWS models calculate exclusion zones (areas where secondary transmis­sions are not allowed/transmit power is set to zero) based on transmitting TV channels and their corresponding tower locations [9].", "sentence2": "we propose a dynamically com­puted exclusion zone characterized as the union of locations where secondary user transmit power must be reduced in order to protect active TV receivers.", "label": "contrasting", "domain": "Networks"}
{"id": "test_989", "sentence1": "That is, the SU network operates fully asynchronously.", "sentence2": "we exploit that the TV transmitter is always transmitting whereas SU-TX transmits intermittently.", "label": "entailment", "domain": "Networks"}
{"id": "test_990", "sentence1": "The required feedback in the previous discussion consid­ers all channels (TV channels, UHF channels) as identical.", "sentence2": "in practice, channels are divided into two types: a physical channel which occupies 6 MHz bandwidth and a vir­tual channel which contains TV programming.", "label": "contrasting", "domain": "Networks"}
{"id": "test_991", "sentence1": "While the broadcasting TV signals and the secondary data are sent in the UHF band, primary feedback can be transmitted out-of-band via WiFi, cellular, or wired connections such as DSL, or in-band via a UHF feedback channel.", "sentence2": "we propose two methods to implement primary feedback with minimum modifications to legacy TV systems: (i) Smart remote: Smartphones can control TVs via infrared, e.g., Samsung Galaxy S5 and HTC One M8.2 smartphones can be used as combined feedback and remote devices.", "label": "reasoning", "domain": "Networks"}
{"id": "test_992", "sentence1": " As in [5], the quadratic form of the internal utility function not only allows for tractable analysis, but also serves as a good second-order approximation for a broad class of concave utility functions.", "sentence2": "ai models the maximum internal demand rate, and bi models the internal demand elasticity factor.", "label": "entailment", "domain": "Networks"}
{"id": "test_993", "sentence1": "A randomly generated code, might be (1) catastrophic, i.e., there exists a non-zero input sequence that can produce all-zero output sequence; or (2) non-equiprobable, i.e., the output values are not uniformly distributed, which can help the ad­versary deploy statistical attacks to distinguish the mapping we aim to conceal.", "sentence2": "the generated code is first validated against above properties, then its free distance d8 is computed.", "label": "reasoning", "domain": "Networks"}
{"id": "test_994", "sentence1": "In the first part of the section, we expose the limitations of the graph method in Euclidean spaces.", "sentence2": "we show, in particu­lar, that conflict graphs do not yield any non-trivial approx­imation to the Scheduling problem in terms of the number of links n. ", "label": "entailment", "domain": "Networks"}
{"id": "test_995", "sentence1": "This analysis shows that one model does not fit all flights.", "sentence2": "the correction step runs whenever GPS data is received.", "label": "neutral", "domain": "Networks"}
{"id": "test_996", "sentence1": "Moreover, to ensure a controlled flight, it is essential to maintain a good quality radio link from the base station to the quadcopter.", "sentence2": "most quadcopter applications do not consider these factors and they tend to use an empirical preset measurement for the maximum distance.", "label": "neutral", "domain": "Networks"}
{"id": "test_997", "sentence1": "Our intention is to run the controller on the quadcopter itself.", "sentence2": "to ease the debugging and testing we run the controller in a laptop in this prototype.", "label": "contrasting", "domain": "Networks"}
{"id": "test_998", "sentence1": "This analysis shows that one model does not fit all flights.", "sentence2": "a model based on the historical data can be used to provide an initial estimation or a hint.", "label": "contrasting", "domain": "Networks"}
{"id": "test_999", "sentence1": "To obtain the accuracy of the localization, an average lo­calization error was calculated.", "sentence2": "the number of the col­lected beacons is directly proportional to the speed of the UAV.", "label": "neutral", "domain": "Networks"}
{"id": "test_1000", "sentence1": "He received his PhD from University of York in 1999.", "sentence2": "david Grace is Head of Communications and Signal Processing Research Group within the Department of Electronics at the University of York.", "label": "neutral", "domain": "Networks"}
{"id": "test_1001", "sentence1": "The rational behind GameOn was that, in many urban cities, commuters spend a large part of their day on a train or a bus.", "sentence2": "it would be great if we availed them of the opportunity to engage in spontaneous mul­tiplayer gaming with their fellow commuters on the same train or bus.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1002", "sentence1": "The next hop of a packet (bird) is defined so that it is (1) repulsed from congested nodes in the repulsion zone of the focal node, (2) attracted to non-congested nodes in the attraction zone of the focal node, (3) and on the way toward the destination.", "sentence2": "we define the following parameters (Figure 2): Z oR of a node is the set that contains its parents and children (one hop).", "label": "reasoning", "domain": "Networks"}
{"id": "test_1003", "sentence1": "We believe that for even more constrained devices, there is no way around specialized, simplified, and highly optimized implementations.", "sentence2": "note that our goal is not to engineer the most memory-efficient network stack but to design a clean, structured, and universal network stack that can be reused for many different IoT use cases, while still being able to cope with constrained environments.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1004", "sentence1": "One category of stacks are ultra-minimalistic implementations, such as the work by Santos et al. [9], which – by design – are not extensible and cannot become a full-featured IP stack. ", "sentence2": "they do not meet the requirements from Section 2.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1005", "sentence1": "Whenever a network event happens, e.g., node failure, link failure or network partition, the event will be detected by the nearby nodes and the update will be propa­gated to every broker within the network domain.", "sentence2": "the brokers can have a consistent view of the network con­dition to help them in deciding proper transmission strate­gies.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1006", "sentence1": "At the same time, there is a considerable amount of traffic from both B and C to D. By duplicating service D at both nodes where B and C reside, we can significantly reduce traffic footprint.", "sentence2": "if the Broker is aware of an intradomain instance of the service, the broker contacts the device that has the instance and requests that it migrates the service to the SEG using a specific transmission module that is suit­able for the underlying network topology.", "label": "neutral", "domain": "Networks"}
{"id": "test_1007", "sentence1": "Service discovery itself is also an important topic that needs further investigation: users could potentially utilise conventional methods such as search engines to discover vari­ous services.", "sentence2": "such services themselves may become inaccessible when the network is isolated.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1008", "sentence1": "We observe that the correlation values between each type of centrality measure computed on the two network layers vary across the various datasets.", "sentence2": "we observe that for SIG-COMM and Social Evolution datasets, characterized by a higher number of nodes (67 and 55, respectively), the correlation between each centrality measure assumes values close to 0, while for the other datasets the correlation varies from weak to strong.", "label": "entailment", "domain": "Networks"}
{"id": "test_1009", "sentence1": "This trend of high variability between the central­ity correlation values of a given dataset can be also observed in UNICAL, LAPLAND and SASSY datasets.", "sentence2": "the interesting result is that across al measures, we note that the degree correlation for UNICAL, UPB, LAPLAND and SASSY, assumes relatively high values, varying from moderate to strong correlation.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1010", "sentence1": "Understanding shopper’s behavior through physical analytics can provide crucial insights to the business owner in terms of e↵ectiveness of promotions, arrangement of products and eciency of services. ", "sentence2": " analyzing shopper’s behavior and browsing patterns is challenging.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1011", "sentence1": "The proposed system does not require the shopper to carry any device as the movements of the shopper is detected purely by observing the variations in CSI of WiFi.", "sentence2": "this mapping between the states and the inferred activities is shown in Tables 1 and 2. ", "label": "neutral", "domain": "Networks"}
{"id": "test_1012", "sentence1": "In addition, there are at most 2k final requests that wish to reach to top row (as each final far request must have entered the tile).", "sentence2": "in total there are at most 3k paths that wish to reach the top side of the NE-quadrant.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1013", "sentence1": "As discussed in Section III-A1, the area of the AND gate is set to be considerably larger than ` the rest (indicated by the symbol).", "sentence2": "the homomorphic operations are implemented using the libScarab cryptographic library [13].", "label": "neutral", "domain": "Networks"}
{"id": "test_1014", "sentence1": "Since there is no other algorithm implementation to be compared, our evaluation consists of the calculation of the frequency that each vehicle changed its state during the simulation.", "sentence2": "we calculate the times that the vehicles join, leave or merge with other vehicles tested in different transmission range scenarios and with varying number of vehicles.", "label": "entailment", "domain": "Networks"}
{"id": "test_1015", "sentence1": "The network topology and the choice of parameters for the benchmark models are motivated by the considerations in previous work [5].", "sentence2": "to isolate the true simulator performance metrics from network model effects such as routing and congestion, we consider network traffic between nearest neighbor nodes only", "label": "entailment", "domain": "Networks"}
{"id": "test_1016", "sentence1": "The most common used techniques are based on L3 and L2 forwarding.", "sentence2": "there are trade-offs in creating a L2 network or L3 network [16].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1017", "sentence1": "The Space-Time Scheduler receives an input set Ji of asso­ciated jobs which contains at least one job.", "sentence2": "the system is stabilized by itself.", "label": "neutral", "domain": "Networks"}
{"id": "test_1018", "sentence1": "(a) Along a straight line (b) Along a curve Figure 8: Traces of moving robot We evaluated the system by executing a swarm application that consists of eleven actions each of which has different spatial-temporal constraints.", "sentence2": "section 5 shows a proof of concept while Section 6 describes the state of the art.", "label": "neutral", "domain": "Networks"}
{"id": "test_1019", "sentence1": "Already in the near future, there will be an enormous amount of wireless devices [14] ranging from deeply embedded sensors and actuators over intelligent gad­gets to fully autonomously acting robots forming sensing and actuating platforms featuring a variety of hardware and software, different programming approaches and system in­terfaces making, especially, the cooperation and coordina­tion a challenging task.", "sentence2": "our approach is to consider all these devices as one emerging system (the swarm) and build a distributed swarm runtime system on top of it that hides heterogeneity and diversity, provides a common in­terface to the outside and decouples swarm applications and their execution in space and time.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1020", "sentence1": "NDN hourglass architecture's thin-waist also supports data exchange instead of service interoperability.", "sentence2": "nDN can be well suited for Open mHealth architecture.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1021", "sentence1": "The key idea of H-CRAN is so called 4C, i.e., clean, centralized, cooperative and cloud, resulting in several notable advantages such as saving CAPEX and OPEX, reducing power consumption, enhancing network capacity, dynamically allocating resources based on current network loads and improving quality-of-experience of mobile subscribers.", "sentence2": "the more complexity that a wireless communication system has, the higher possibility with which security vulnerabilities will occur.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1022", "sentence1": "In the wireless access network, radio signals are transmitted in the open shared medium, typically the air environment.", "sentence2": "they are easily eaves­dropped by adversaries.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1023", "sentence1": "The environ­ment is special in underground coal mine, with a high incidence of accidents[3].", "sentence2": "the monitoring in coal mine monitoring is becoming increasingly important.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1024", "sentence1": "In this paper, the number of base station is limited, but the positioning accuracy achieved is very impressive. ", "sentence2": "it is validated that the positioning algorithm and the model is very consistent with the requirements of location in coal mine.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1025", "sentence1": "The rest of the paper is organized as follows.", "sentence2": "the system through­put and the number of selected RDs grow with the increase of p0.", "label": "neutral", "domain": "Networks"}
{"id": "test_1026", "sentence1": "However, if the decoding process of MAC header fails, no noti.cation signal will be generated and broadcasted.", "sentence2": "the transmit­ter (Alice) stops transmitting and all the nodes, including clients and AP, restart the channel contention immediately.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1027", "sentence1": "Note that, the pt-channel contention and transmission are proceeded concurrently and transparently with AP transmission.", "sentence2": "process for pt-channel contention and transmission should be completely collision-free for AP transmission.", "label": "entailment", "domain": "Networks"}
{"id": "test_1028", "sentence1": "We exam­ine the effectiveness and advantages of LIPS in Section 5.", "sentence2": "we also omit it due to the limited space.", "label": "neutral", "domain": "Networks"}
{"id": "test_1029", "sentence1": "We try to present fast algorithms for these theories, and improve the perfor­mance of our coding scheme.", "sentence2": "fig. 9(a)-(d) plot the information throughput improvement return on the occupancy of decoding resource queue.", "label": "neutral", "domain": "Networks"}
{"id": "test_1030", "sentence1": "To support a multi-hop D2D communication (even intra-cell), the routing and scheduling algorithms must be carefully designed.", "sentence2": "we employ the statistics of the previous flow information to calibrate the topology.", "label": "neutral", "domain": "Networks"}
{"id": "test_1031", "sentence1": "Device-to-device (D2D) communication is envisioned as a promising technique to supplement the cellular network, which can enhance the spectral efficiency, extend the cov­erage, and create new services.", "sentence2": "the current D2D communication only supports one-hop communication with­in one cell.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1032", "sentence1": "For typical operating conditions, the residual self-interference is above the noise-floor, and this can make it diffcult to demodulate weak external signals.", "sentence2": "for cognitive-radio, the goal is only to detect the presence of a primary user, and not necessarily to correctly decode the transmissions.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1033", "sentence1": "This allows us to reuse the receiver design from the prover for the verifier, which greatly simplified the implementation.", "sentence2": "both verifier and receiver shared the same design and most components.", "label": "entailment", "domain": "Networks"}
{"id": "test_1034", "sentence1": "We agree with the general intuition of the principle in [9], but would reformulate it in the following way: Minimize the length of the symbol used to present a single bit as well as the processing time the transceiver needs for demodulating a single bit challenge and sending the response bit.", "sentence2": "distinguish the two possible trans­mitted bit values at the receiver within as short a time as is feasible and send an answer as fast as possible.", "label": "entailment", "domain": "Networks"}
{"id": "test_1035", "sentence1": "The delay itself is not a prob­lem regarding accurate distance measurements: As long as it is known, distances can still be computed with high preci­sion.", "sentence2": "a malicious prover may exploit high expected processing delays to shorten measured distance by replying early, for instance using a sophisticated receiver structure with lower receive processing delay.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1036", "sentence1": "Advancing only the preamble (without advancing the data pulses) also advances the sampling intervals, leading to invalid received replies.", "sentence2": "the attacker cannot advance the preamble more than its processing advantage.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1037", "sentence1": "In this section, we will discuss a transceiver design for a UWB rapid-bit-exchange phase.", "sentence2": "we will study which modulation scheme is best suited to implement rapid-bit-exchange given a fixed bandwidth.", "label": "entailment", "domain": "Networks"}
{"id": "test_1038", "sentence1": "That allows an attacker to shorten the ranging result between two honest nodes.", "sentence2": "the attacks are possible due to the large integration window (128 ns) of the receiver.", "label": "entailment", "domain": "Networks"}
{"id": "test_1039", "sentence1": "To simplify analysis, we restrict our attention to deterministic environments.", "sentence2": "we can assume the environment is modeled as synchronous deterministic automaton that receives the nodes' ack outputs as input, and generates their bcast inputs as its output.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1040", "sentence1": "For example, a 3-uniform jamming adversary may jam channel one for group one, jam channel one and six for group two, and do nothing for group three.", "sentence2": "an n-uniform jamming adversary can make jamming decision for each node in­dividually.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1041", "sentence1": "In the lemma below, we show a lower bound for winning the (c, k)-bipartite hitting game.", "sentence2": "we first calculate the player s losing probability.", "label": "entailment", "domain": "Networks"}
{"id": "test_1042", "sentence1": "There­fore this assumption could be dropped if one is not interested in implementing an absMAC that performs local broadcast exactly in G1-e.", "sentence2": "our absMAC implementation outputs rcv events for all bcast-messages received, which can be modified if required by other higher-level algorithms designed for absMACs and when the G1-e-neighborhood is known.", "label": "entailment", "domain": "Networks"}
{"id": "test_1043", "sentence1": "As a network module in Linux, tc does not reference Linux kernel's time as the way user application does.", "sentence2": "tc is transparent to our virtual time system.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1044", "sentence1": "It is also important to note that the deployments vary from relatively simple ones such as the epilepsy case study, to complicated ones such as the incontinence and depression case studies.", "sentence2": "in the case studies we show that using the framework we can quickly implement and deploy multimodal, largely passive behavioral monitoring systems that are useful to caregivers and medical professionals.", "label": "entailment", "domain": "Networks"}
{"id": "test_1045", "sentence1": "The process takes about 20 minutes.", "sentence2": "the modules that we have built, available in a library, are meant to be cross-platform and have been tested on a variety of platforms such as the Raspberry Pi, Beaglebone, and laptops.", "label": "neutral", "domain": "Networks"}
{"id": "test_1046", "sentence1": "We have developed software (based on the Arduino programming language [2]) uploaded on network nodes, tailored to the role and functionality of each node.", "sentence2": "the software executed by microcontrollers undertakes the collection of information measured by sensors, data processing and information transmission to the wireless medium.", "label": "entailment", "domain": "Networks"}
{"id": "test_1047", "sentence1": "By processing it, we obtained that there were 88 smart meters, 3 of them working as switches, with 4, 3 and 2 nodes reaching the concentrator through each of them.", "sentence2": "as in many other communications standards, PRIME specification just sets the maximum values of certain communications parameters that may influence overall network performance, thus leaving them up to the manufacturer.", "label": "neutral", "domain": "Networks"}
{"id": "test_1048", "sentence1": "Finally, we address the second question asked at the beginning this section, evaluating the reconfiguration latency of the Blueswitch design for variable number of rules.", "sentence2": "the flow-table pipeline sends the matched output actions to per-port packet marshallers via an 1-to-N arbiter.", "label": "neutral", "domain": "Networks"}
{"id": "test_1049", "sentence1": "It does not rely on specialized drivers like the other frameworks and can be used with every NIC as long as this card is supported by Linux.", "sentence2": "without modi­fied drivers the NIC cannot push the packets directly to user space.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1050", "sentence1": "The next drop in the graph is the transition between L3-cache and non-cached RAM accesses.", "sentence2": "even under these circumstances the model provides a good estimation if the average per-packet costs are used.", "label": "neutral", "domain": "Networks"}
{"id": "test_1051", "sentence1": "A recent survey indicated that the number of L7 appliances (e.g., middleboxes) in SDN’s infrastructures is comparable to the number of routers [29], making high layer processing in SDN highly challenging.", "sentence2": " to empower SDN with such functions, high layer visibility and protocol parsing in the device are necessary. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1052", "sentence1": "We translate DCCFG into DCRG in a similar way.", "sentence2": "the growth of DCCFGs does not linearly increase the unique leaf nodes because of the very similarities between multiple DCCFGs.", "label": "neutral", "domain": "Networks"}
{"id": "test_1053", "sentence1": "The PI-based approaches benefit from the linear-complex prior DFA on the input and achieve much faster goodput than SP-based ones.", "sentence2": "it is also affected by the RegEx libraries.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1054", "sentence1": "Encrypted traffic increases continuously in today’s Internet for better security performance, which obstructs the finegrained identification on the packets.", "sentence2": "we argue that the content parsing is still necessary and feasible for SDN.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1055", "sentence1": "But the opposite is not always true, x might match an ELCP of a range that does not contain x.", "sentence2": "on top of ORange1 we present the ORange-k classiffication scheme that supports k dimensional ranges and achieves exponential with k space improvement by using cascaded instances of the ORange1 classifier.", "label": "neutral", "domain": "Networks"}
{"id": "test_1056", "sentence1": "First, an OCS does not process packet headers and requires a separate control plane to configure optical circuits before data can flow through the interconnect.", "sentence2": "we also discuss scheduling ineffcien­cies of current fast OCSs and evaluate ideas to solve them.", "label": "neutral", "domain": "Networks"}
{"id": "test_1057", "sentence1": "There are huge tradeoffs between these metrics, and it is easy to achieve high values in one of them sacrificing one of the others.", "sentence2": "latency and throughput are often conflicting goals (e.g., larger batch sizes improve throughput at the expense of latency).", "label": "entailment", "domain": "Networks"}
{"id": "test_1058", "sentence1": "Using this tree structure, the root node of each Sync Tree holds the combined hash of all the names stored under the represented collection.", "sentence2": "we describe and evaluate a functional prototype that executes in CCNx-compatible environments.", "label": "neutral", "domain": "Networks"}
{"id": "test_1059", "sentence1": "Unlike the IP paradigm, NDN focuses on what the content is and not on where the content is located.", "sentence2": "in the NDN architecture, a packet is delivered according to its name rather than its destination address.", "label": "entailment", "domain": "Networks"}
{"id": "test_1060", "sentence1": "If more than one consumer asks for the same data, the PIT entry of the requested name will contain the faces of all the requesting consumers.", "sentence2": "on its way back, the Data packet will be forwarded to all the requested faces.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1061", "sentence1": "All other (internal) nodes of an address represent XIDs, and each node is associated with between one and four strictly prioritized outgoing edges; four being the maximum fanout supported in XIP addresses.", "sentence2": "also, XIA zFilter cleanly interoperates with TCP/IP, as §6.1 demonstrates.", "label": "neutral", "domain": "Networks"}
{"id": "test_1062", "sentence1": "Fast packet forwarding Since the maximum fanout in an XIP address is four edges, an XIA router may have to inspect up to four XIDs to make a routing decision.", "sentence2": "even without hardware optimizations, Linux XIA can efficiently instantiate XIA s forwarding mechanism by not serializing these lookups.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1063", "sentence1": "However, k hash functions might generate fewer than k different hash values in practice.", "sentence2": "we proactively compute sufficient extra hash values for each address, which we refer it as internal collision avoidance.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1064", "sentence1": "Particularly, our results show small filters require k hash values of an address to be different in contrast to large filters.", "sentence2": "hash functions might generate fewer than k different hash values in practice.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1065", "sentence1": "Caesar's focus is on very small filters, which is different from a similar usage of the core property in previous work [31].", "sentence2": "our results show small filters require k hash values of an address to be different in contrast to large filters.", "label": "entailment", "domain": "Networks"}
{"id": "test_1066", "sentence1": "Recent studies [8, 6] have predicted that the number of connecting devices and active address prefixes will jump to 50 billion and 1.3-2.3 million, respectively, by the end of 2020.", "sentence2": "the current rapid growth of the number of address prefixes (i.e., about 17% per year) is the root of many existing problems for operators, who have to continuously shrink the routing and forwarding tables of their devices or upgrade to increasingly more expensive data planes [13].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1067", "sentence1": "The backup path is built from the fast memory and handles uncommon cases where the primary path is not reliable due to false positives in the filters thus rarely is less efficient when it accesses the RIB (Sec 4).", "sentence2": "the primary path ensures the common-case high-speed forwarding while the backup path guarantees the correctness.", "label": "entailment", "domain": "Networks"}
{"id": "test_1068", "sentence1": "Moreover such hardware features have helped devel­opers write efficient IDS systems that now achieve processing performances that are comparable with commercialized hardware solutions.", "sentence2": "nIDSes based on such hetero­geneous systems (GPU-based) [3, 4] do have some drawbacks: it is normally difficult to program a SIMD module on a GPU such that it delivers peak performance while keeping the power consumption and processing latencies in check.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1069", "sentence1": "For 512k flow entries, the total DRAM space needed is 6.5 MB.", "sentence2": "the memory space requirements for these tables can be accommodated in the SRAM and DRAM available on board.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1070", "sentence1": "The sketch data structure is capable of summarizing high-speed network traffic in real-time and producing accurate estimation result of a specific traffic measurement task.", "sentence2": "due to the fixed hardware implementation of the sketch counter array, the flexibility of adopting the same data structure for various traffic measurement applications is limited.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1071", "sentence1": "However, in the multiple flow tables, a final action stored in the Action Table is cascaded by the multiple middle actions of the rules.", "sentence2": "one action of a rule cannot be found in the Action Table anymore.", "label": "entailment", "domain": "Networks"}
{"id": "test_1072", "sentence1": "We also assume that the attacker knows all positions of all nodes at all times.", "sentence2": "it knows the exact positions of the verifiers at the reception of its location claims already when sending those claims.", "label": "entailment", "domain": "Networks"}
{"id": "test_1073", "sentence1": "In an indoor environment, many users use WLAN technologies such as IEEE 802.11 access points to connect to the Internet.", "sentence2": "iEEE 802.11 technologies operate on an ISM band, which has limited bandwidth, and experience interference from nearby access points and other devices.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1074", "sentence1": "Note that when channel condition is good, our proposed scheme shows similar performance with the legacy one.", "sentence2": "as the channel condition deteriorates and increases transmission failures, the proposed scheme divides large packets into smaller ones and thereby increases the number of successful transmission.", "label": "neutral", "domain": "Networks"}
{"id": "test_1075", "sentence1": "In recent years, future Internet architecture proposals have re­visited the idea of source routing for its benefits while addressing its scalability and security problems.", "sentence2": "the scalability problems have been addressed by using source-based path selec­tion, a system in which the source obtains a list of paths that it can use to reach the destination, rather than a full topology map as in generic source routing schemes.", "label": "entailment", "domain": "Networks"}
{"id": "test_1076", "sentence1": "To improve the performance of cell edge users, the system has to allocate more resource to cell edge users, and hence the resource allocated to the inner users decreases, and thus, the sum throughput decreases.", "sentence2": "we can control the efficiency and fairness by appropriately adjusting the QoS requirement.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1077", "sentence1": "Especially, in this heterogeneous network (HetNet), the interference to each user increases sharply, since the num­ber of entities transmitting simultaneously increases as the number of cells is increases.", "sentence2": "the enhanced intercell interference coordination (eICIC) is considered as one of key schemes to reduce interference in the HetNet.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1078", "sentence1": "First and foremost, we thank the many engineers at Aka­mai who designed, implemented and rolled-out end-user map­ping, making it possible for us to evaluate its impact.", "sentence2": "we present our experience and insights obtained in de­ploying end-user mapping for clients around the world in the first half of 2014.", "label": "neutral", "domain": "Networks"}
{"id": "test_1079", "sentence1": "Western Europe sees low distances appearing in a small band.", "sentence2": "korea and Taiwan are significant in having the smallest distances.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1080", "sentence1": "A total of 37,294 ASes with the most demand were analyzed.", "sentence2": "the cached resolution is only valid for the IP block for which it was provided and not for any client IPs that do not belong to the block.", "label": "neutral", "domain": "Networks"}
{"id": "test_1081", "sentence1": "However, if the client cluster has a large radius, i.e., the clients are far away from each other, there may be no single server assignment for the entire cluster that is optimal for 1000 all clients in it.", "sentence2": "it is inherently difficult for NS-based mapping to perform well when the client cluster has a large 0 radius, even knowing client-LDNS pairings.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1082", "sentence1": "One heuristic approach to reducing the number of mapping units for end-user mapping is to use the IP blocks (i.e., CIDRs) in BGP feeds that are the units for routing in the Internet.", "sentence2": "if a set of /24 IP blocks belong within the same BGP CIDR, these blocks can be combined since they are likely proximal in the network sense. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1083", "sentence1": "The second server then serves the content to the client.", "sentence2": "this process incurs a redirection penalty that is acceptable only for larger downloads such as media files and software downloads.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1084", "sentence1": "The reason is that the mapping system could discover the client cluster and assign servers that provide good performance for the entire cluster.", "sentence2": "if the client cluster has a large radius, i.e., the clients are far away from each other, there may be no single server assignment for the entire cluster that is optimal for all clients in it. ", "label": "contrasting", "domain": "Networks"}
{"id": "test_1085", "sentence1": "It is clear that EU provides a large benefit over the other schemes for higher percentiles of ping latency.", "sentence2": " NSbased mapping provides diminishing benefits beyond 160 deployment locations for the 99th percentile latency, and is in particular unable to reduce it below 186 ms even with 1280 deployment locations.", "label": "entailment", "domain": "Networks"}
{"id": "test_1086", "sentence1": "Our measurement study of client-LDNS distances in Section 3 is based on a much wider global cross-section of clients and LDNSes than prior work and largely confirm prior conclusions on public resolvers.", "sentence2": "different scoring functions that incorporate bandwidth, latency, packet loss, etc can be used for different traffic classes (web, video, applications, etc).", "label": "neutral", "domain": "Networks"}
{"id": "test_1087", "sentence1": "In fact, both CDNs and content providers are focused on improving the performance of the worst-performing client.", "sentence2": "we computed the 95th and 99th percentiles of the latencies, i.e, latencies for 1-5% of the worst clients.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1088", "sentence1": "δ represents a tradeoff between safety and efficiency; larger values of δ yield alibis who are so far away from a forbidden region that, if packets were to traverse both, there would be a very large increase in latency over a normal path through the alibi. ", "sentence2": "with a large δ, one may be less likely to find a viable alibi, but that alibi is likely to work even in the face of variable round-trip times and congestion.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1089", "sentence1": "The goal of provable avoidance routing is detection, as opposed to prevention.", "sentence2": "alone, it is unable to ensure a user s packets will not traverse a region of the world we do not require modi.cations to the underlying routing protocols or hardware, and so we are subject to all of today s uncertainties as to where packets will travel.", "label": "entailment", "domain": "Networks"}
{"id": "test_1090", "sentence1": "Consider for example a peer who has a satellite link with extremely high latencies: such a peer may never satisfy the alibi conditions, regardless of the path his or her packets take.", "sentence2": "a benevolent relay whose packets never traverse the forbidden region might never be viable simply because it has poor connectivity.", "label": "entailment", "domain": "Networks"}
{"id": "test_1091", "sentence1": "This obviously has higher costs than the previous solution, but it might be the only possibility in those areas where the density of home APs is not sufficient, or where the buildings are too far from the street.", "sentence2": "operators can use hardware of better quality for these dedicated APs, thereby providing increased throughput and extended coverage to their users.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1092", "sentence1": "The first possibility is to leverage existing wireless access points (APs) that are nowadays installed in almost every home, office, or caf´e.", "sentence2": "this means sharing the physical capacity of the 802.11 radio channel(s) and the available bandwidth of the Internet connection between a private and a public Wi-Fi network.", "label": "entailment", "domain": "Networks"}
{"id": "test_1093", "sentence1": "The Android API is used to execute the algorithm choice, basing on previous measurement for particular location.", "sentence2": "there is a trade-off between frequent updates and the amount of transferred data, hence the update rate should be chosen wisely.", "label": "neutral", "domain": "Networks"}
{"id": "test_1094", "sentence1": "There exist also several solutions based on Kalman filter formulation [1, 2, 8, 9], but the analysis is only based on simulations without assessing their implementation feasibility for sensor nodes.", "sentence2": "the solution proposed in [6] is simple enough to be implemented on sensor nodes but lacks the accuracy to meet the ever increasing synchronization demands of WSN applications.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1095", "sentence1": "For these rules, existing algorithms are still inefficient in terms of classification speed.", "sentence2": "the number of required bits is always prohibitively high, making this solution impractical due to the memory footprint of the 2l buckets.", "label": "neutral", "domain": "Networks"}
{"id": "test_1096", "sentence1": "Packet classification is required to achieve high throughput while fitting into a commodity memory hierarchy.", "sentence2": "fewer memory accesses and a reasonable memory footprint are the main concerns when designing a packet classification algo­rithm.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1097", "sentence1": "Experiments show that ESCC (only leave trace in routers two hops from clients), has already achieved better performance than the universal caching, in term of access delay, link traffic and server load reduction (Fig.3-5).", "sentence2": "eSCC could be a substitute for the costly universal caching.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1098", "sentence1": "Public Key Infrastructures and thus also the RPKI are complex systems.", "sentence2": "current tools to inspect these objects are based on command line interfaces.", "label": "neutral", "domain": "Networks"}
{"id": "test_1099", "sentence1": "A recently deployed example is prefix origin validation, which authorizes an AS to announce an IP prefix.", "sentence2": "table 1 shows the performance gain in terms of rsync calls and download time thanks to reduced communication overhead.", "label": "neutral", "domain": "Networks"}
{"id": "test_1100", "sentence1": "MultiSpot charges different types of devices simultaneously.", "sentence2": "we need a second equation that determines the relationship between the voltage one applies to the transmitter, VT and the resulting transmitter current, IT .", "label": "neutral", "domain": "Networks"}
{"id": "test_1101", "sentence1": "The figure shows that initially, the phone closer to the transmitter (i.e., Rx1) charges faster since it has a stronger magnetic cou­pling with the charger.", "sentence2": "once this phone is fully charged, MultiSpot transfers the power to the second receiver (Rx2), in­creasing its charging rate.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1102", "sentence1": "With this increased efficiency, it may be possible to charge a receiver at a larger distance when more receivers are in the system.", "sentence2": "this experiment is aimed to find the maximum distance from the transmitter a phone can still charge from, given a number of re­ceivers.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1103", "sentence1": "Unfortunately, in practice, one has to use a voltage source instead.", "sentence2": "we need a second equation that determines the relationship between the voltage one applies to the transmitter, VT and the resulting transmitter current, IT .", "label": "reasoning", "domain": "Networks"}
{"id": "test_1104", "sentence1": "However, even when the number of multi-paths changes due to environmental changes, the speed of change in lengths of multi-paths does not change because it depends on the movement of human body and not on the number of multi-paths in an environment.", "sentence2": "the frequency components in the CFR power stay the same as long as the person performs the same activity, and DWT gives higher energy in the same levels regardless of how many multi-paths have appeared or disappeared.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1105", "sentence1": "Furthermore, once CARM is trained on the given training set, it can be directly applied to environments and persons that have not been included in the training set.", "sentence2": "cARM does not need the on-site training data collection as for E­eyes [27].", "label": "reasoning", "domain": "Networks"}
{"id": "test_1106", "sentence1": "For example, running and falling both have similar frequencies but the duration of falling is shorter than running.", "sentence2": "to analyze CFR power for human activities, we need to extract frequencies from it at multiple resolutions on multiple time scales.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1107", "sentence1": "Several network management tasks require fine grained traffic classification.", "sentence2": "a number of techniques have been proposed over the years to address this problem; a survey of these may be found in [9, 18].", "label": "reasoning", "domain": "Networks"}
{"id": "test_1108", "sentence1": "Cov and max stand for the ratio of apps identified by SAMPLES rule set and all identifiable apps in the test set.", "sentence2": "the length of the hash divided by the cardinality of the app executable archive repository.", "label": "neutral", "domain": "Networks"}
{"id": "test_1109", "sentence1": "The hash for identifier type app id contains within it the app id as key, and a constant 1 as the corresponding value.", "sentence2": "the hash for app name contains the app name as the key, and the corresponding app id as its value.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1110", "sentence1": "Through experiments over a vast corpus of mobile applications (over 700 K), we have demonstrated the comprehensiveness and applicability of our approach across all three platforms.", "sentence2": "the coverage is significantly higher than the other state-of-the-art methodologies from literature.", "label": "entailment", "domain": "Networks"}
{"id": "test_1111", "sentence1": "Nowadays most of metropolitan areas are overloaded with WiFi APs, many of which are private.", "sentence2": "it is still a waste of energy to blindly wake up a WiFi NIC when an arbitrary AP shows up, without knowing if it is accessible or not.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1112", "sentence1": "We see that different tags have different response delays and bit durations, which result in bit misalignment and non-overlapped state transition boundaries.", "sentence2": "the response delay offset and the bit duration offset (normalized by period) can be up to 30% and 1% for each bit, respectively.", "label": "entailment", "domain": "Networks"}
{"id": "test_1113", "sentence1": "To investigate channel coefficient distributions in practice, we first characterise the backscatter channel of multiple tags using the SDR testbed.", "sentence2": "the WISP tags are programmed to backscatter known preambles and payloads, so we can directly iden­tify the states of all the tags in each symbol cluster.", "label": "entailment", "domain": "Networks"}
{"id": "test_1114", "sentence1": "Around 40% of cases have 0 bit errors in BiGroup when four tags transmit concurrently, and around 20% have an average BER below 0.05 in BiGroup when five tags transmit concurrently.", "sentence2": "the low­est average BERs when four and five tags collide for the LA based scheme are much higher (>0.1 BERs for all collisions).", "label": "contrasting", "domain": "Networks"}
{"id": "test_1115", "sentence1": "The RNC provides control functions and radio resource management for the cellular base stations (NodeBs).", "sentence2": "the dynamics of each chunk include a high bandwidth initial burst that quickly exceeds the downlink RLC buffer threshold, triggering a promotion to DCH.", "label": "neutral", "domain": "Networks"}
{"id": "test_1116", "sentence1": "Obviously, the common post denoising process not only can­not recover the original video, but also deteriorate the video quality compared to the watch-only version in all five basic metrics due to recognizing noise incorrectly.", "sentence2": "kALEIDO is insuscepti­ble to common denoising attack, and guarantees the reliable video privacy preservation.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1117", "sentence1": "Notice that the recording schemes of smart devices are typically designed to mimic the basics of how human views the surrounding world.", "sentence2": "there is a very small design opportunity for such a transformative technology.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1118", "sentence1": "The usual pirate videotaping scenario would be worse than this ideal testing scenario.", "sentence2": "if the quality of the pirate video in this ideal scenario is intolerable, the pirate video taken in worse conditions will experience more severe quality degradation.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1119", "sentence1": "Additionally, unstable inter-frame intervals of most commercial onboard cameras aggravate the information loss and distortion for both Case 1 and Case 2, hence making the color distortion of recorded frames even worse.", "sentence2": "during a single capture time window (Tc), rolling shutter effect and unstable intervals could cause a temporal information loss or deviation in the recorded frame if the displayed frames for each original video frame are time-varying. ", "label": "reasoning", "domain": "Networks"}
{"id": "test_1120", "sentence1": "The average keystroke recognition ac­curacy rate for user 10 in previous experiment, which used 30 samples for training classifiers was just 80%.", "sentence2": "we can conclude that increasing the number of training samples increases the accuracy of WiKey.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1121", "sentence1": "For the features extracted for keys 'I' and 'o' shown in figures 5(a) and 5(b), the DTW distance among features of key 'I' was 18.79 and the DTW distance among features of key 'o' was 19.44.", "sentence2": "the average DTW distance between features of these keys was 44.2.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1122", "sentence1": "For an ADC sampling rate of 20 kHz and modulation window of 128 points, a light beacon lasting for 6.4 ms is sufficient for the photodiode to separate light rays.", "sentence2": "the overhead of transmitting light beacons is negligible given that a data packet typically lasts for hundreds of ms based on the IEEE 802.15.7 standard [8].", "label": "reasoning", "domain": "Networks"}
{"id": "test_1123", "sentence1": "Consider a single photodiode on the floor, we hypothesize that if any opaque object stands in the direct path between the point light source and the photodiode, the photodiode will not be able to perceive any light coming from this point light source.", "sentence2": "the photodiode observes a light intensity drop compared to the case when there is no object blocking its direct path to the light source.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1124", "sentence1": "The design of light beacons is driven by the observation that while the perceived light intensity represents the sum of all incoming light rays, these light rays can be separated in the frequency domain if they flash at different frequencies.", "sentence2": "if we transform a time series of perceived light intensity to the frequency domain using Fast Fourier Transform (FFT), we can observe frequency power peaks at the frequencies at which these light rays flash.", "label": "entailment", "domain": "Networks"}
{"id": "test_1125", "sentence1": "Hence, an LED light Li flashing at frequency fi leads to not only a global power peak (main frequency power) at frequency fi, but also small local power peaks at all the harmonics frequencies (Figure 5(c)). ", "sentence2": " if the perceived light intensity from light Li changes, it will affect not only the main frequency power at f, but also the power peaks at harmonics. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1126", "sentence1": "The nice property of the FI is that the forwarders (C) are sending the symbols that the previous nodes (B) along the path have already seen.", "sentence2": "the interfering symbols are known.", "label": "entailment", "domain": "Networks"}
{"id": "test_1127", "sentence1": "The reason FI cancellation works is based on the fact that the signal of FI is correlated with the output signal from the same radio so that FI can be predicted.", "sentence2": "this does not hold for all the interference components, like the one shown in Figure 2(b).", "label": "contrasting", "domain": "Networks"}
{"id": "test_1128", "sentence1": "To deliver data direct from this radio, Y is a better choice.", "sentence2": "when the radio is a forwarder and the channel to its ancestor is bad, delivering the packet alone in that link takes 200µs.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1129", "sentence1": "Realizing in-band cut-through in wireless networks, however, is challenging in many ways as we will discuss later.", "sentence2": "wireless cut-through not only improves delay performance but also improves end-to-end throughput.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1130", "sentence1": "Typical DASH protocols [5,8,10] use the throughput observed from each video segment to estimate the available end-to-end network bandwidth.", "sentence2": "such estimations stay blow the bandwidth unless the segment size (and equivalently the video bitrate) is large enough to saturate the network pipeline. ", "label": "contrasting", "domain": "Networks"}
{"id": "test_1131", "sentence1": "When the buffer level is low (we empirically set the threshold to 2000ms), even if a rare mistake can incur intolerable buffer underrun.", "sentence2": "the bandwidth (esti­mated by the PIRS component) is considered to jump to another level when the highest video bitrate below the band­width (i.e., the quantized bandwidth) changes.", "label": "neutral", "domain": "Networks"}
{"id": "test_1132", "sentence1": "A re­cent world-wide measurement survey [2] reveals that, even in regions with wide LTE coverage, LTE only increases video quality by 20% over 3G.", "sentence2": "the average stalling time remains 7.5 to 12.3 seconds for each minute s mobile video playback [2].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1133", "sentence1": "For CSI collection, there is a fundamental gain limitation of 1 because CSI consists of only information about the link between one antenna and another antenna.", "sentence2": "signals received at other antennas do not contain information about that link's CSI.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1134", "sentence1": "Since these down link sequences need to be detected prior to synchronization, they must have low streaming auto-correlations, both with themselves and the other sequences in the orthogonal set.", "sentence2": "since the sequences must be detectable without knowledge of when they start, the receiver must perform a full correlation at every sample, thus a time-shift of the sequences must produce a low correlation; otherwise it could cause an erroneous detection.", "label": "entailment", "domain": "Networks"}
{"id": "test_1135", "sentence1": "In the low RSSI regime we see that the 64-length beacon begins to perform poorly, and is only able to correct 80% of the beacons to within 2 kHz error.", "sentence2": "the 128-length beacon with low RSSI is performs similarly to the high RSSI 64-length, which indicates extending the beacon length could further reduce CFO estimation error.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1136", "sentence1": "Our second key insight is that synchronization and association are not time-critical.", "sentence2": "synchronization is valid for 100s of ms and association only happens once; thus by reducing the frequency of synchronization Faros is able to substantially reduce the channel overhead of these operations in the no-CSI mode, at the cost of slightly increased association latency at the cell edges.", "label": "entailment", "domain": "Networks"}
{"id": "test_1137", "sentence1": "LTE already provides a compelling random access solution which fits well within the Faros design, with the exception that Faros allows for longer length sequences to be employed to finely tune the gain gap.", "sentence2": "with each users path delay to estimate timing advance, with small computational overhead.", "label": "neutral", "domain": "Networks"}
{"id": "test_1138", "sentence1": "Users which send reference signals in a given resource element gain spatial resource elements in the corresponding time and frequency coherence interval for both the uplink and downlink phases.", "sentence2": "any given reference symbol provides an estimation that is valid both for the coherence time interval, as well as a wider frequency coherence interval.", "label": "entailment", "domain": "Networks"}
{"id": "test_1139", "sentence1": "To obtain CSI, the transmitter sends a pre-known sequence, called a pilot, which the receiver uses to compute this amplitude and phase shift for each subcarrier.", "sentence2": "this requires time-frequency synchronization, as without time synchronization the receiver would not reliably know where the pilot starts, and without frequency synchronization there would be inter-subcarrier interference that causes inaccurate channel estimation", "label": "contrasting", "domain": "Networks"}
{"id": "test_1140", "sentence1": "In an 802.11-like CSMA MAC, the beacon would need to be replaced by the beamformed Faros beacon immediately followed by the CSI collection phase, including the dedicated random ac­cess and association slots.", "sentence2": "cDD spreads the power output of all M antennas spatially, and can be thought of as arbitrarily beamforming on different subcarriers.", "label": "neutral", "domain": "Networks"}
{"id": "test_1141", "sentence1": "Through our study, we observe that for any three consecutive WiFi bands, if the CSI phases in the first and the third channels are well compensated, they can serve as two anchors to further cali­brate the CSI phase from the second band if its phase error is not fully corrected.", "sentence2": "we can rotate the CSI phase from the second band and stop when its phase differences for the overlapped sub-carriers with other two bands are minimized.", "label": "entailment", "domain": "Networks"}
{"id": "test_1142", "sentence1": "Our evaluations show that the localization accuracy can be substantially improved.", "sentence2": "splicer improves the CUPID localization accuracy by 71%, with median localization errors about 0.95m.", "label": "entailment", "domain": "Networks"}
{"id": "test_1143", "sentence1": "The reason is that offsets are frequency independent.", "sentence2": "we find that the CSI phase accuracy can be further improved by leveraging the primarily spliced result", "label": "neutral", "domain": "Networks"}
{"id": "test_1144", "sentence1": "We plan to adapt this threshold with varying SNR values in our future work.", "sentence2": "a single threshold is performing pretty well as explained.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1145", "sentence1": "ToneTrack makes mistakes only in the very low SNR region (below 6 dB). ", "sentence2": " ToneTrack leverages the insight that we can sometimes still retrieve useful and relative accurate information from these inaccurate pseudospectra.", "label": "neutral", "domain": "Networks"}
{"id": "test_1146", "sentence1": "Compare the relative amplitudes of the two peaks.", "sentence2": "the smaller reflection path peak has a much larger error, so we can still extract relatively accurate information from the MUSIC spectrum in these scenarios as we only care the directpath peak. ", "label": "neutral", "domain": "Networks"}
{"id": "test_1147", "sentence1": "µACK in [58] can also harvest such a throughput gain with the help of extra hardware (additional antennas working on another frequency band).", "sentence2": "in Figure 17, we further evaluate two approaches in three typical 802.11 WLANs environments: a open-space academic hall, a large library, and an indoor of.ce, where the frequency selective fading levels increase.", "label": "neutral", "domain": "Networks"}
{"id": "test_1148", "sentence1": "Figure 4(b) shows that error events usually contain ed or ed + 1 coded bit errors for all the code rates.", "sentence2": "ed and ed + 1 together account for more than 80% of the error events for all four code rates.", "label": "entailment", "domain": "Networks"}
{"id": "test_1149", "sentence1": "The first two applications aim to evaluate the accurate PER prediction of Recitation.", "sentence2": "the former one focuses on the PER prediction when PER is close to zero and the latter one focuses on the prediction with the PER transition range.", "label": "entailment", "domain": "Networks"}
{"id": "test_1150", "sentence1": "The vast majority of service issues are rapidly detected via the network and/or application.", "sentence2": "figure 7h shows the CDF of the loss ratio of detected failures during quiet hours (i.e., low usage period) and busy hours (i.e., peak usage period).", "label": "neutral", "domain": "Networks"}
{"id": "test_1151", "sentence1": "Moreover, approximated users' location could be used to localize geographical location of a failure to benefit root cause analysis.", "sentence2": "as shown in Figure 6, ABSENCE consists of four components: historical usage retrieval, hourly usage retrieval, time series processing and user location profile retrieval.", "label": "neutral", "domain": "Networks"}
{"id": "test_1152", "sentence1": "Thus, one could natively assume that we could look for drops in load on network elements to identify service outages.", "sentence2": "where in the network to look for such reductions in carried load is an intriguing challenge - traffic is regularly shifting around a mobile network, typically without any service impact.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1153", "sentence1": "However, users might still be impacted, even when the RAN accessibility KPI is good, because users that lost radio coverage are not even accounted for in the KPI calculation or when the root cause of the service problems are beyond the RAN (e.g., congestion on a core network element). ", "sentence2": "service disruption occurs when accessibility is bad, but accessibility being good does not necessarily imply that service is good.", "label": "entailment", "domain": "Networks"}
{"id": "test_1154", "sentence1": "However, if the transmitter is a client device, the WiDrawenabled receiver may have to periodically probe the device to obtain its AoA values.", "sentence2": "wiDraw may only be able to leverage a few neighboring clients.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1155", "sentence1": "MUSIC can compute both the azimuth and elevation of an AoA if the receiver s anten­nas are arranged in a special pattern [12].", "sentence2": "when applied to linear antenna arrays such as those within a laptop or tablet screen, MUSIC can only compute a one-dimensional representation of the AoA (.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1156", "sentence1": "If large environmental variations affect an AoA near the hand's trajectory, this will reflect in our trajectory tracking error.", "sentence2": "if such an AoA is far away from the hand's trajectory, it needs to be detected as an outlier.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1157", "sentence1": "Expectedly, as the word gets longer, it is more difficult to recognize it correctly.", "sentence2": "even by using a simple handwriting recognition software such as MyScript, the average word recognition accuracy is more than 91%.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1158", "sentence1": "If the transmitter is an AP, WiDraw can obtain the CSI and AoA of its transmitted signal by leveraging periodic beacons.", "sentence2": "if the transmitter is a client device, the WiDraw enabled receiver may have to periodically probe the device to obtain its AoA values.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1159", "sentence1": "Researchers have analyzed the performance of Multipath TCP in such hybrid networks [7, 1, 5, 2].", "sentence2": "exploring Mobile/WiFi Handover with Multipath TCP.", "label": "neutral", "domain": "Networks"}
{"id": "test_1160", "sentence1": "Thereby, MPTCP follows the end-to-end principle and does not require changes to the network or the application, i.e., an MPTCP connection appears like a regular TCP connection to the network.", "sentence2": "many approaches, such as Mobile IP [5] and application layer solutions, were proposed, but are not widely used.", "label": "neutral", "domain": "Networks"}
{"id": "test_1161", "sentence1": "We set the alpha threshold to 0.05 (i.e., the model should describe 95% of the data), and we find that indeed the p-value is always lower than alpha, validating that multi­ple linear regression can properly .t our data, and therefore our multiple linear regression results can be trusted.", "sentence2": "we do that so to determine if 802.11ac is a standard that could possibly support low latency show con­trol systems in the future (Fig.", "label": "neutral", "domain": "Networks"}
{"id": "test_1162", "sentence1": "Second, operators can perform active measurements with client devices to probe the network.", "sentence2": "these samples can only provide a rough estimation of the network performance, as they only cover the geo-temporal-span of the active measurements.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1163", "sentence1": "Therefore, it is of great importance to stimulate user participation so as to obtain high-quality data.", "sentence2": "simply providing monetary rewards is not feasible because rational users can untruthfully report information of real costs, arrival and departure time, for maximizing their benefits.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1164", "sentence1": "Figure 1 shows some of the potential indoor application scenarios at 60 GHz.", "sentence2": "60 GHz links are highly vulnerable to propagation loss.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1165", "sentence1": "Lei Wang is the corresponding author.", "sentence2": "compared with indoor localization schemes, CLaWa is a lightweight solution which does not require expensive localization operations or supporting devices.", "label": "neutral", "domain": "Networks"}
{"id": "test_1166", "sentence1": "The utilization of in­termittent networks will become increasingly important with the explosion of mobile data traffic.", "sentence2": "use of the ex­tremely high frequency bands to augment the mobile access network remains a work in progress.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1167", "sentence1": "The JUJU services together with Metal As a Service (MAAS) are able to program physical severs on-demand allowing for programmability of cloud resources (computing, storage, and networking) and RF hardware in the remote radio head.", "sentence2": "jUJU and MAAS can be seen as key enablers behind the so called programmable cloud, which adapts the infrastructure according to the requirements of slices instantiated (e.g., by deploying an appropriate number of adapted cloud servers with the LXC virtualization).", "label": "reasoning", "domain": "Networks"}
{"id": "test_1168", "sentence1": "If the shortest path between a pair of sites is longer in the proximity graph than in the Delaunay triangulation, it indicates miss­ing edges, or a lack of connectivity in the proximity graph (which may not be a problem).", "sentence2": "one can infer geographic information from handover behavior.", "label": "neutral", "domain": "Networks"}
{"id": "test_1169", "sentence1": "From such data one can infer that if a user appears in two different cells within a short span of time, that a handover took place, and that the coverage areas of the two cells overlap.", "sentence2": "one can infer geographic information from handover behavior.", "label": "entailment", "domain": "Networks"}
{"id": "test_1170", "sentence1": "Many testbeds for WSNs have been proposed with the goal of enabling users to test new methods and algorithms in a realistic environment.", "sentence2": "many of these testbeds have only been developed for indoor use [8, 4, 7, 3].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1171", "sentence1": "In case the device has not received the chunk yet by the end of the dissemination period, it will make a new query to the Azure Mobile Service to request the chunk.", "sentence2": "at least one query and up to two queries per chunk are performed.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1172", "sentence1": "In this work, we focus on the design and implementation of solutions that address the problem of video content delivery in practice -by far the most challenging in terms of network resources among the set of use cases above.", "sentence2": "our ideas and concepts can be applied or reformulated for other use cases as well.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1173", "sentence1": "Each chunk that has not be fully downloaded due to commu­nication failure is lost.", "sentence2": "the size of the chunk should be sufficiently small to guarantee an efficient delivery, e.g. in presence of users with short contact duration δ.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1174", "sentence1": "Our aim is not to develop a new estimator, but fusing link quality with the routing metrics already in place in opportunistic routing protocols.", "sentence2": "we are interested in the message replication, so for each message (with a given ID) we measure the time between replication on a particular node.", "label": "neutral", "domain": "Networks"}
{"id": "test_1175", "sentence1": "We run a simulation scenario re.ecting our broader re­search interest in the use of opportunistic networks for sen­sor data collection, specifically for challenging environments such as after a natural disaster.", "sentence2": "we assume con­strained low-power wireless nodes with transmission char­acteristics of those seen for typical wireless sensor nodes.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1176", "sentence1": "To test out this hypothesis, and to gain insights into ways of providing effective content delivery services in public transportation systems, we analyze a dataset collected from an Internet service provider in public transport buses in Sweden.", "sentence2": "traditional content delivery mechanisms, such as using caches within the network is not effective in the context of public transportation systems.", "label": "neutral", "domain": "Networks"}
{"id": "test_1177", "sentence1": "Safari browser accounts for 86% of the iOS traffic.", "sentence2": "ma­jority of the Android traffic is divided between the Chrome Mobile (55%) and the Android Browser (33%).", "label": "contrasting", "domain": "Networks"}
{"id": "test_1178", "sentence1": "Since contact opportunities may take an arbitrarily long time to occur, nodes, who often have limited storage capabilities, may have to store data for long periods of time and network congestion can quickly build up.", "sentence2": "dTN con­gestion control is critical to ensure nodes are congestion-free and, when needed, can serve as relays for messages to be delivered end-to-end.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1179", "sentence1": "This increases the number of relayed messages in the network considerably while the number of delivered messages remains the same which increases the overhead.", "sentence2": "it increases the amount of network resources that are used to deliver one message to its destination.", "label": "entailment", "domain": "Networks"}
{"id": "test_1180", "sentence1": "This increases the number of relayed messages in the network considerably while the number of delivered messages remains the same which increases the overhead.", "sentence2": "the SPMBM [9] model uses Dijkstra's shortest path algorithm to calculate the shortest path from the current location to a randomly selected destination.", "label": "neutral", "domain": "Networks"}
{"id": "test_1181", "sentence1": "A Bluetooth scatternet can thus link together more than 8 devices.", "sentence2": "the notion of macronet is wider than that of scatternet, as the micronets forming a macronet can use different communication technologies.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1182", "sentence1": "This is supported by the large Pearson correlation coefficient presented.", "sentence2": "selection of DCU deployment sites must evaluate the site's ability to support minimum data volume requirements.", "label": "neutral", "domain": "Networks"}
{"id": "test_1183", "sentence1": "Sharing opportunities are sparse but exist.", "sentence2": "we observe one pair of home APs, node 0 and 7, which exhibit recip­rocal sharing relationships.", "label": "entailment", "domain": "Networks"}
{"id": "test_1184", "sentence1": "Prior work has studied optimiz­ ing the AP placement to maximize the signal strength in certain areas [16].", "sentence2": "moving the AP to improve the strength in one area would also result in the decline of signal strength in another area.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1185", "sentence1": "Each molecule, during its trip, can hit the spherical receiver with radius rr and, in this case, the hitting molecule is absorbed by the spherical receiver.", "sentence2": "a molecule can contribute to the signal just once.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1186", "sentence1": "As for the boundary conditions of the shell, we use the partially inelastic model.", "sentence2": "when a nanoparticle collides with the shell from both inside and outside the shell (both spherical and cylindrical), we handle a partially elastic collision with the plane that is tangent to the shell in the collision point.", "label": "entailment", "domain": "Networks"}
{"id": "test_1187", "sentence1": "For an aperture of 90°, the performance begins to depend on the radius of the shell, even if still not providing any gain.", "sentence2": "for small values of the shell radius, the effect influences the absorption process, and improves with R.", "label": "entailment", "domain": "Networks"}
{"id": "test_1188", "sentence1": "It appears that most of absorptions occur in the front side of the receiver.", "sentence2": "our idea is to put a shell to trap the carriers which have passed beyond the receiver, but still between the shell and the receiver itself.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1189", "sentence1": "Based on the SNRs, we see quite expected error performance curves, with the best performance represented by the highest scattering case due to the largest received energy.", "sentence2": "even with the largest transient, the error quickly approaches the boundary of 0.5.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1190", "sentence1": "The power delay profile describes the power transfer in the channel.", "sentence2": "it gives a power impulse response for a signal.", "label": "entailment", "domain": "Networks"}
{"id": "test_1191", "sentence1": "In the previous work, we used log-normal size distribution for the diameter of the scattering particles.", "sentence2": "there are commercially available small particles, which have desired features for possible scattering measurement campaigns, including controlled spherical shape with known mean diameter and standard deviation.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1192", "sentence1": "The delayed energy also potentially increases the probability of inter-symbol interference (ISI).", "sentence2": "we leave the ISI analysis to the future work.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1193", "sentence1": "The multiple scattering helps the detection of the symbols due to higher bit energy at the receiver.", "sentence2": "this exposes the system to potential ISI if the symbol separation is not sufficiently large.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1194", "sentence1": "It lays its eggs inside a particular kind of mussel.", "sentence2": "endangered fish habitat monitoring Sensors We installed five types of sensors (water temperature, air temperature, dissolved oxygen (DO), illuminance, and humidity) into the biotope.", "label": "neutral", "domain": "Networks"}
{"id": "test_1195", "sentence1": "In Figure 8 (1), the RTT is evaluated for different ICMP packet payloads and for different distances.", "sentence2": "the policy concerning the TCP sender window is examined for direct link and multi-hop network scenarios.", "label": "neutral", "domain": "Networks"}
{"id": "test_1196", "sentence1": "In other words, Kbest is equivalent to a global optimal solution in IHS under the conditions that the problem has a global optimal solution.", "sentence2": "hA assumes that Kbest is a local optimal solution vector.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1197", "sentence1": "It can be seen that IHS converges after about 3800 iterations and SA doesn t arrive in a stable state.", "sentence2": "but the illumination requirement is left out of account.", "label": "neutral", "domain": "Networks"}
{"id": "test_1198", "sentence1": "We can take advantage of IHS to find the regions of the solution area and increase the diversity of solution vectors in the HM.", "sentence2": "respectively, the dynamic update method of PAR and BW are discussed in details in [7].", "label": "neutral", "domain": "Networks"}
{"id": "test_1199", "sentence1": "When the light beam is blocked by human body, the existing pulse will disappear in frequency domain.", "sentence2": "by detecting the change in frequency domain, for each pair of LED light and photodiode on the floor, the system will know whether the light beam is blocked. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1200", "sentence1": "To address the above limitations, in this paper, we seek a dif­ferent approach to human identification.", "sentence2": "we examine the feasibility of using the ubiquitous light around us to recognize who we are, without requiring users to carry any on-body devices, neither using any cameras constantly monitoring users.", "label": "entailment", "domain": "Networks"}
{"id": "test_1201", "sentence1": "In [4], computer vision and triangulation is used to estimate the relative location between several lights and the camera.", "sentence2": "our future works include enhancements in the following aspects.", "label": "neutral", "domain": "Networks"}
{"id": "test_1202", "sentence1": "A lens is placed in front of the LCD to project light going through different pixels of the LCD toward different spatial locations.", "sentence2": "different spatial locations on the same plane would receive different positioning signals, which allows the receiver to estimate its own location with respect to the transmitting light.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1203", "sentence1": "The testbed capabilities are demonstrated through multiple graphical user interfaces (GUIs) at the relay, destination, and trace recorder nodes.", "sentence2": "in each relay node we plot the instantaneous SINR and queueing information, while a GUI at the laptop-PC controlling the destination node and primary receiver (Fig.", "label": "entailment", "domain": "Networks"}
{"id": "test_1204", "sentence1": "The DTS packet is used to announce the information update on code, transmit power, and queueing to neighboring nodes.", "sentence2": "we build a 7-node software-defined radio testbed and imple­ment a distributed algorithm that maximizes the secondary network throughput, while at the same time avoids inter­ference to primary users through joint Routing and cOde­waveform CHannelization (ROCH).", "label": "neutral", "domain": "Networks"}
{"id": "test_1205", "sentence1": "A tradeo↵ exists between the CPU cost and the simulation accuracy in terms of received messages.", "sentence2": "the primary nodes locations can be changed during the demo execution time.", "label": "neutral", "domain": "Networks"}
{"id": "test_1206", "sentence1": "The pendula are custom designed wooden devices operated via servo motors, which we will operate using our Z1/TelosB nodes.", "sentence2": "time synchronization provides the basis for several appli­cations in wireless sensor networks but the limited memory and computational power, and the use of low precision oscil­lators make the task of time synchronization non-trivial.", "label": "neutral", "domain": "Networks"}
{"id": "test_1207", "sentence1": "The initial trust establishment, or Device Pairing, cannot rely on physical interactions between unfamiliar de­vice holders.", "sentence2": "such a receipt is, and only is, verifiable to A, B and BS using their private keys, while the obfuscated feedback is verifiable to the receiver without revealing its actual value.", "label": "neutral", "domain": "Networks"}
{"id": "test_1208", "sentence1": "The emerging Device-to-Device (D2D) communication tech­niques [2] such as WiFi Direct, Bluetooth Smart, and LTE direct have a great potential to provide cost-e.ective solu­tions to collaborative interactions among mobile devices for mobile crowdsourcing [3, 4].", "sentence2": "since participating devices are private and temporarily hired, there is no prior trust relationship between devices opportunistically encoun­tered.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1209", "sentence1": "They promise unprecedented new abilities to observe and understand large-scale, real-world phenomena at a fine spatio-­temporal resolution.", "sentence2": "this potential does not come for free; WSNs have been known to be vulnerable to several types of attacks aiming at compromising their security.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1210", "sentence1": "The analysis involves several network characteristics and is aimed towards identifying easily measured features that can be used for efficiently detecting and classifying DoS attacks in WSNs.", "sentence2": "this potential does not come for free; WSNs have been known to be vulnerable to several types of attacks aiming at compromising their security.", "label": "neutral", "domain": "Networks"}
{"id": "test_1211", "sentence1": "Figures 5-7 depict the BER performance of the DF scheme described above under Rician fading.", "sentence2": "each figure corresponds to a different Rician k-factor (0, 10 and 20 respectively).", "label": "entailment", "domain": "Networks"}
{"id": "test_1212", "sentence1": "In this topology, the wireless links between both transmitters and the relay are considered ideal, i.e. the SNR value of the frames received by the relay is very high and the frames are always decoded correctly.", "sentence2": " the SNR value of the frames received by the receiver from the relay and the two transmitters varies.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1213", "sentence1": "In hypervisor approach to virtualization (i.e. VM), IO virtualization is done through the hardware emulation layer under the control of hypervi­sor, where as in container this is done through the device mapping.", "sentence2": "direct access to the hardware is easier in containers than in VMs as they operate at the host OS level.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1214", "sentence1": "The evaluation results in Section 3 confirm that uplink process­ing dominates the downlink, and that the total processing increases with PRB and MCS.", "sentence2": "the contribution of each underly­ing BBU functions to the total processing time and how they scale with the increase of PRB and MCS remains to be analysed so that an accurate model could be build.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1215", "sentence1": "ROCH capabilities are demonstrated through multiple graphical user interfaces (GUIs) at the relay (R1, R2), destination (D), and trace recorder (TR) nodes.", "sentence2": "gUIs at both secondary relay nodes illustrate real-time SINR and queueing information, while the laptop-PC, that controls the secondary destination and primary receiver (PRx) nodes, compares the instantaneous throughput between the primary link (PTx– PRx) and the secondary two-hop network. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1216", "sentence1": "All these works consider WPANs, where all clients are located in a single room, the traffic pattern consists mainly of client-to-client connections, clients themselves can act as relays, and each client maintains a connection to the AP.", "sentence2": "abstracting with credit is permitted.", "label": "neutral", "domain": "Networks"}
{"id": "test_1217", "sentence1": "In order to recognize these patterns, their colors must be distinguishable, even in cases of low light, hardware noise, and ambient light.", "sentence2": "there must be a noticeable color transition between different patterns.", "label": "entailment", "domain": "Networks"}
{"id": "test_1218", "sentence1": "For example, the grayscale value of 100 might normally be considered black and representing 0 in QR Codes, but for ImplicitCode, it could be 1 if the carrier video block is even darker. ", "sentence2": "we also need to determine the threshold between the colors representing 1 or 0 relative to the carrier video, even though we cannot assume any a priori knowledge of the carrier video. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1219", "sentence1": "Across all four datasets, WiScan consistently achieves 90%+ of the optimal connectivity, demonstrating that our SSID learning scheme accurately predicts future SSIDs.", "sentence2": "wiScan achieves 97% of the optimal in the Canberra dataset.", "label": "entailment", "domain": "Networks"}
{"id": "test_1220", "sentence1": "Overestimating the scan interval will make the system miss potential connectivity until the next scan.", "sentence2": "we configure scan interval Ts conservatively.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1221", "sentence1": "There is only a single point for WiScan in Figure 9 because WiScan automatically adapts it scan interval based on user mobility and AP distribution.", "sentence2": "wiFisense and existing method in Android work with a fixed scan interval, so we configure their fixed scan interval to different values and examine their tradeoff between energy consumption and achieved connectivity.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1222", "sentence1": "endpoint congestion over a long period of time.", "sentence2": "copyrights for components of this work owned by others than ACM must be honored.", "label": "neutral", "domain": "Networks"}
{"id": "test_1223", "sentence1": "These speculative packets travel on a low-priority virtual channel (VC) [12], and are only al­lowed a limited queuing time inside the network before be­ing dropped by an otherwise lossless network.", "sentence2": "hPC networks may not always be dominated by large mes­sage transfers.", "label": "neutral", "domain": "Networks"}
{"id": "test_1224", "sentence1": "Overall, varying the number of indirect routes considered at each packet injection (nI ) affects slightly the throughput saturation point under worst-case traffic, with higher numbers providing better results, as more available routes are available.", "sentence2": "cSF affects the average delay under high loads of uniform traffic. ", "label": "contrasting", "domain": "Networks"}
{"id": "test_1225", "sentence1": "All end-nodes of RA send their traffic to the end-nodes of RC and all nodes from RB send to all nodes of RD.", "sentence2": "the link that connects RB and RC becomes overloaded, with 2p flows passing per direction, resulting to 1/2p of the maximum throughput.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1226", "sentence1": "If Rook causes unlikely values to occur more frequently then there could be a shift to a more uniform frequency distribution versus the normal traffic. ", "sentence2": "it might be that the velocity of an avatar is never actually greater than 100.0.", "label": "neutral", "domain": "Networks"}
{"id": "test_1227", "sentence1": "The authors do not show that the availability subproperty is met. ", "sentence2": "the authors did not evaluate the impact of address rotation on the network infrastructure near the hosts.", "label": "entailment", "domain": "Networks"}
{"id": "test_1228", "sentence1": "Future work could integrate these mechanisms into the DNS capabilities system itself to better distinguish legitimate users from adversaries.", "sentence2": "attackers may avoid using the ISP DNS infrastructure as­sociated with each compromised system to avoid detection when performing large-scale scanning.", "label": "entailment", "domain": "Networks"}
{"id": "test_1229", "sentence1": "We then created another application, fpanaly.py [21], to analyze the results.", "sentence2": "as an example, consider these TCP options (which happen to come from a SonicWALL firewall device): NOP, NOP, MSS=1440, SAckOK, NOP, WScale=0 ", "label": "neutral", "domain": "Networks"}
{"id": "test_1230", "sentence1": "We then created another application, fpanaly.py [21], to analyze the results.", "sentence2": "we wanted to determine which probes produced the highest amount of variability in the responses.", "label": "entailment", "domain": "Networks"}
{"id": "test_1231", "sentence1": "KL-divergence also requires large data sampling and dense fingerprints in signal distribution comparison.", "sentence2": "it cannot adapt to the noisy airport environment.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1232", "sentence1": "Moreover, the signal coverage of APs in our campus corridor and HKCP is constrained by the wall partitions, which helps differentiate the RPs.", "sentence2": "we expect a better localization performance on HKUST and HKCP than in HKIA.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1233", "sentence1": "We first show the metrics’ abilities to accurately interpret the probe reception history.", "sentence2": "given the history, we observe how precisely a metric captures the performance of the corresponding link.", "label": "entailment", "domain": "Networks"}
{"id": "test_1234", "sentence1": "Based on the observations in the previous section we design cETX on top of the Gilbert Model. ", "sentence2": "this model is only capable of representing the status of a single link.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1235", "sentence1": "In this work, we observe that spatial correlation interplays with temporal correlation to exhibit reception correlation at consecutive links.", "sentence2": "there exists spatiotemporal correlation, i.e., the packet reception of a link is dependent on the recent reception at preceding links along a path.", "label": "entailment", "domain": "Networks"}
{"id": "test_1236", "sentence1": "Immediate forwarding is likely to exhibit spatiotemporal correlation.", "sentence2": "when the delay is sufficiently large, cETX reflects the inde-e2 e4 pendence between consecutive links (i.e., no spatiotemporal correlation) where cETX still remains more accurate than ETX due to its consideration on temporal correlation.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1237", "sentence1": "The packet reception ratios of Link3 and Link4 are 0.49 and 0.5, leading to approximately the same ETX of 2 for both links.", "sentence2": "the difference in the true average transmissions is quite large; 2.05 and 2.98 for Link3 and Link4, respectively.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1238", "sentence1": "End-to-End Delivery Ratio: The success probability of a packet delivered end-to-end.", "sentence2": "figure 19 shows the end­to-end delay of metrics in the three protocols.", "label": "neutral", "domain": "Networks"}
{"id": "test_1239", "sentence1": "End-to-End Delivery Ratio: The success probability of a packet delivered end-to-end.", "sentence2": "the number of (re)transmissions in each link in a path do not exceed the retransmission limit.", "label": "entailment", "domain": "Networks"}
{"id": "test_1240", "sentence1": "To further maintain fairness, both metrics use the history of up to 90 probes, the value suggested by ETX [11].", "sentence2": "when there is a data packet to be forwarded (every 10s in our case), both metrics use the history of the last 90 probes to estimate the number of (re)transmissions until a successful delivery.", "label": "entailment", "domain": "Networks"}
{"id": "test_1241", "sentence1": "(ii) Spatiotemporal correlation captured via data traffic in cETX naturally reflects the forwarding scheme adopted by the network.", "sentence2": "the corresponding degree of spatiotemporal correlation is captured regardless of immediate or delayed (e.g., duty cycled networks) packet forwarding.", "label": "entailment", "domain": "Networks"}
{"id": "test_1242", "sentence1": "The flattening phase still presents room for improvements: increasing its efficiency is largely a matter of e.ective engi­neering.", "sentence2": "the solving phase is difficult to further optimize, as the problem is inherently NP-hard.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1243", "sentence1": "The goal of our design is to find the largest idle intervals of the microcontroller for Fountain coding, under the precondition of controlling the synchronization offset between the concurrent transmissions of constructive interference within 0.5 µs.", "sentence2": "we need to guarantee that the generated idle CPU intervals are constant and the coding computation occupies deterministic time.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1244", "sentence1": "During the dissemination, the power consumption of Pando is the same as the previous protocols, since all of them keep the nodes in active mode.", "sentence2": "compared with the previous protocols, benefiting from a shorter dissemination time, Pando reduces the energy consumption of individual nodes by the same reduction factor as the dissemination time.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1245", "sentence1": "As in Figure 5, the leaf nodes only transmit in their even TX slots, corresponding to the oddRXslotsoftheirparentnodes, inwhichtheparent nodes listen to the feedback from the leaf nodes.", "sentence2": "the silence of leaf nodes in the time slots of odd packets does not impact the feedback process.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1246", "sentence1": "Unlike CAs, IP PKGs need not to interact with users during the signed object verifying procedures.", "sentence2": "once private keys are obtained, the sIP address holders will no longer necessary to communicate with the IP PKGs, thus signif­icantly reducing the communication and transmission cost while implementing routing veri.cation procedures.", "label": "entailment", "domain": "Networks"}
{"id": "test_1247", "sentence1": "Besides, consecutive requests may also impact content delivery in terms of added queueing and processing delay which we assume to be roughly proportional to the number of messages.", "sentence2": "the cost can increase faster and requests may be dropped in a busy network.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1248", "sentence1": "It is worth noting that many topological properties (e.g., average degree, density etc.) are homogeneous on random networks.", "sentence2": "a randomly chosen sub-network possesses similar characteristics as the whole network which is also known as self-similarity.", "label": "entailment", "domain": "Networks"}
{"id": "test_1249", "sentence1": "The reason is that the optimal radii of the nodes with high ( values in both static and dynamic flooding are small and close to each other.", "sentence2": "dynamic flooding usually significantly increases the radius of the nodes with low B.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1250", "sentence1": "A server managed by an administrator is considered to be a trustable message exchange point among related peers or a kind of portal site for this SNS that is known to everybody in the serving network.", "sentence2": "the SNS administrator can observe any message sent to the public or to a specific friend group.", "label": "entailment", "domain": "Networks"}
{"id": "test_1251", "sentence1": "Therefore, in NDN, security is built into contents, rather than connections between end hosts.", "sentence2": "content publishers must sign all of their contents, i.e., any content in NDN has a signature to prove its provenance and integrity.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1252", "sentence1": "Originator-signed signatures inherent in Information Centric Networking assure the in­tegrity and provenance of messages exchanged among peers, which makes it possible to realize moderator-controlled in­formation sharing in which a peer can become a moderator and control the distribution of his private message group as a trustable server.", "sentence2": "moderated content requires mul­tiple signatures, which increases the size of the exchanged message and is inadequate, especially for short message ser­vices.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1253", "sentence1": "Boneh et al. proposed the concept of aggregate signatures in [4], in which it is possible to aggregate n signatures on n distinct messages from n distinct users into a single aggregated signature of constant size.", "sentence2": "(5) The moderator sends the content to corresponding subscribers or peers within his group.", "label": "neutral", "domain": "Networks"}
{"id": "test_1254", "sentence1": "Current NDN testbed key management [3] uses a simple hierarchical trust model with a root key, which signs the keys for each site.", "sentence2": "this NDN’s default pull-based communication model can also be extended to have pushbased multicast capability (pub/sub), as demonstrated in COPSS [6].", "label": "neutral", "domain": "Networks"}
{"id": "test_1255", "sentence1": "For example, passive probing cannot detect link recovery, that is, a QoS improvement of a path which is not among the current working paths; it is hard to determine the performance of an interface that does not send or receive data.", "sentence2": "active probing can probe interfaces other than the current working path to find recovered or better interfaces.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1256", "sentence1": "One key difference to our approach is that they use routing to locally address content: peers add known content chunks to their FIB and the consumer application performs a longest prefix match on the chunk name, which returns only the face of the local peer.", "sentence2": "our system relies mainly on forwarding: we assume that the FIB contains multiple interfaces per prefix (prefixes can be more coarse-grained than chunk-level) and let the forwarding strategy decide which one to use.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1257", "sentence1": "For that, NDN has a built-in security mechanism based on digital signature.", "sentence2": "resultingly, the content is verified repeatedly at each insertion(more precisely, when a cache-hit occurs after each insertion).", "label": "neutral", "domain": "Networks"}
{"id": "test_1258", "sentence1": "As concerns the cache hit rate, a smaller value of p results in a higher hit rate when s = 0.7.", "sentence2": "as the content popularity becomes more skewed, the benefit of probabilistic caching disappears, and CBS achieves lower hit rates than the proposed scheme.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1259", "sentence1": "This decoupling of content from location enables efficient content distribution by minimizing redundant transmission on the links. ", "sentence2": "it brings about new types of security problems as well.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1260", "sentence1": "Especially, if poisoned contents lie in the network cache, called content store(CS), interests would be served by the poisoned content rather than they propagate toward the content server.", "sentence2": "users whose interests pass through the contaminated CS cannot access the valid content.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1261", "sentence1": "Like in the TCP/UDP case, the continuous space representation captures the aggregate flow behavior neglecting microscopic packet-level dynamics.", "sentence2": "the smaller the packet size and the larger the number of multiplexed flows/packets, the more accurate is the fluid approximation of system dynamics.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1262", "sentence1": "First, our analysis is based on the simplifying assumption of a perfect allocation of IO name registrations into BF con­tainers, while the actual number of BFs depends on the merging process throughout the hierarchy (see Section 3).", "sentence2": "we consider the selection of a BF configuration among the ones satisfying Equation 5 for s = S.", "label": "neutral", "domain": "Networks"}
{"id": "test_1263", "sentence1": "However, to overcome the high path loss inherent at these high frequencies, mm­wave networks must employ highly directional beamform­ing antennas, which makes link establishment and mainte­nance much more challenging than in traditional omnidi­rectional networks.", "sentence2": "maintaining connectivity under node mobility necessitates frequent re-steering of the transmit and receive antenna beams to re-establish a direc­tional mm-wave link.", "label": "entailment", "domain": "Networks"}
{"id": "test_1264", "sentence1": "Finally, the related issue of neighbor discovery in directional wireless ad-hoc networks has been widely discussed in the literature [21, 18, 9, 13, 4].", "sentence2": "these works focus on fast discovery of multiple directional links between all pairs of nodes in the network, rather than the fast establishment of the individual direc­tional links between one node pair.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1265", "sentence1": "Other schedulers do not give priority to the faster vehicles; hence, they may experience larger delays re­sulting in larger position error.", "sentence2": "the SLA scheduler gives the priority to faster vehicles, and therefore prevents large errors for faster vehicles, resulting in greater control of the position error of the vehicles.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1266", "sentence1": "As LTE is an infrastructure-based network, vehicles cannot broadcast their safety messages directly to their neighbors; thus, all messages should pass through the infrastructure.", "sentence2": "these messages may congest the network and lead to high delays for some vehicles.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1267", "sentence1": "There remain a number of outstanding issues to address in order for the co-operative scheduling protocol to achieve the robustness required for use in wireless LANs.", "sentence2": "a mechanism should be added to detect and recover from a lost schedule packet.", "label": "entailment", "domain": "Networks"}
{"id": "test_1268", "sentence1": "The affected nodes then need a way to return to normal scheduling.", "sentence2": "once we have the scheduling architecture in place, the Intent protocol proceeds as follows.", "label": "neutral", "domain": "Networks"}
{"id": "test_1269", "sentence1": "We note that the results of the fixed case (without cell zoom­ing) is same in any scenarios because it keeps the complete coverage regardless of user distribution until all batteries run out.", "sentence2": "the BS cell is a circle with a radius rang­ing from 0m to 400m with 20m steps.", "label": "neutral", "domain": "Networks"}
{"id": "test_1270", "sentence1": "The usage situation of these transmission lines is essential information for car­rying out the simulation at each terminal.", "sentence2": "it is essential for the protocol for carrying out carrier sensing, such as CSMA/CA.", "label": "entailment", "domain": "Networks"}
{"id": "test_1271", "sentence1": "Computer sim­ulations are used because they can be conducted compar­atively easily.", "sentence2": "simulations using inexpensive computers have come into wide use due to the improved per­formance of personal computers in recent years.", "label": "entailment", "domain": "Networks"}
{"id": "test_1272", "sentence1": "Since the frequency of memory access increases due to increases in the number of threads, there is some in­crease in calculation time due to memory bus competition, but compared with a CPU, a GPU is almost unaffected by an increase in the number of terminals.", "sentence2": "when the number of terminals was 1024, it was confirmed that the GPU can calculate at a speed about 4.6 times greater than the CPU.", "label": "entailment", "domain": "Networks"}
{"id": "test_1273", "sentence1": "In particular, simulations using inexpensive computers have come into wide use due to the improved per­formance of personal computers in recent years.", "sentence2": "calculation time and precision depend greatly on computer performance, and if the simulation scale increases due to fac­tors such as a large number of sensor terminals, the amount of calculation needed to obtain high-precision results will be tremendous, and it will take an extremely long time to obtain results.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1274", "sentence1": "If the simulation scale is large and the number of terminals will not fit in the same thread block, then transmission lines can be realized using global memory.", "sentence2": "in this case the memory access latency will become extremely large, and as a result, there will be a major increase in the time needed for the simulation.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1275", "sentence1": "In general, even though real-time image processing places an extremely high load on hardware, the nature of the processing itself is comparatively simple, and thus is suited to implementation in hardware.", "sentence2": "nVIDIA Corpo­ration and other GPU developers are competing to improve performance through higher level integration of high-speed memory and powerful processing capabilities.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1276", "sentence1": "The cluster coefficient measures the tendency of nodes to cluster on the graph.", "sentence2": "we can measure the density of connections among the neighbors in the graphs.", "label": "entailment", "domain": "Networks"}
{"id": "test_1277", "sentence1": "The IEEE 802.11 standard [6] does not have collision detection, and so the protocol places more importance on avoiding collisions in the first place.", "sentence2": "when a node with a frame to send senses the medium transition from busy to idle it does not deterministically start its own transmission and instead a node waits a random backoff.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1278", "sentence1": "In vehicular networks, the packet losses in the path between source and destination occur due to packet collisions, fast movement and a limited communication range of vehicles.", "sentence2": "when a node with a frame to send senses the medium transition from busy to idle it does not deterministically start its own transmission and instead a node waits a random backoff.", "label": "neutral", "domain": "Networks"}
{"id": "test_1279", "sentence1": "As a consequence, the total amount of static overhead is N H .", "sentence2": "figure 5: Fraction of Monitored Vehicles 2 different vehicular densities in the district of Manhattan, NY, and in Rome.", "label": "neutral", "domain": "Networks"}
{"id": "test_1280", "sentence1": "In previous work it has been shown that the proposed mechanisms are capable of improving CoAP-based end-to-end communications in IoT networks [2, 3].", "sentence2": "the investigations of advanced congestion control mechanisms so far only have covered reliable CoAP communications, not taking into account unreliable communications, such as the ones based on CoAP observe.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1281", "sentence1": "Xiang et al. [5] developed an algorithm for SVC flow adjustment for video streaming over wireless networks using DASH (Dynamic Adaptive Streaming over HTTP) [6].", "sentence2": "before the simulation phase, it is also necessary to generate two additional traces.", "label": "neutral", "domain": "Networks"}
{"id": "test_1282", "sentence1": "VANETs face particular challenges such as fast topology changes, low link lifetime or a potentially high number of nodes taking part in the network, among others.", "sentence2": " the two former issues have encouraged researches to propose geographical routing protocols for VANETs that make their routing decision based only on local information and that do not need to construct end-toend paths.", "label": "entailment", "domain": "Networks"}
{"id": "test_1283", "sentence1": "The simulated results comprise the energy needed by the radio transceiver and the microcontroller of each node in the network.", "sentence2": "simulation was carried out (i) without timing annotation, i.e., ignoring execution times of firmware, (ii) including execution times of low-level drivers, and (iii) including all firmware runtimes.", "label": "neutral", "domain": "Networks"}
{"id": "test_1284", "sentence1": "[13] made a strong case for including application code in the simulation, as the complexity of firmware used in modern WSNs is increasing.", "sentence2": "to the best of our knowledge, we are the first showing the impact in terms of energy consumption of ignoring firmware runtimes in WSN simulations.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1285", "sentence1": "sQualNet provides good scalability, accurate battery models, and also models the clock drift of nodes.", "sentence2": "the integrated data logger enables long­running measurements.", "label": "neutral", "domain": "Networks"}
{"id": "test_1286", "sentence1": "Despite the considerable improvements achieved by IG and FIG algorithms compared to other benchmarks, several issues can be found in the algorithms: 1) each UE greedily maximizes its immediate payoff based on its neighbour UEs’ strategy of the last iteration. ", "sentence2": "the quality of the solution is highly sensitive to the UE update sequence; many such sample paths may lead to bad local optima.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1287", "sentence1": "Compared to IG and FIG, RL and FP make more intelligent decisions on strategy updates by exploiting historical information.", "sentence2": "the experiment setup is as follows.", "label": "neutral", "domain": "Networks"}
{"id": "test_1288", "sentence1": "But no matter what content is in rm, the FC can still give the same result as long as the number of malicious SU's false reports is below the aggregation's threshold which is usually set as half the number of SUs [20].", "sentence2": "if less than half SUs are malicious, our scheme's correctness can be kept.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1289", "sentence1": "All the parties in the semi-honest model must follow the protocol of collaborative sensing and our scheme, but they can keep their own intermediate results, and in this model, honest users are the majority [15].", "sentence2": "sUs and the FC must honestly do coin flipping and send their result whether they are semi-honest or not.", "label": "entailment", "domain": "Networks"}
{"id": "test_1290", "sentence1": "Generally, we randomly choose a SU as the Helper for one round before the sensing starts.", "sentence2": "for example, a malicious Helper simply drops all reports from SUs and sends a mess to the FC, then the sensing process cannot finish as expected.", "label": "neutral", "domain": "Networks"}
{"id": "test_1291", "sentence1": "In that case, varying the channel too quickly creates a lot of over­head.", "sentence2": "the channel still needs to change often enough to avoid long delays.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1292", "sentence1": "Based on this novel approach, we develop service process characterizations for fading channels with finite blocklength channel coding, leading to novel probabilistic delay bounds that can give a fundamental insight into the capabilities and limitations of wireless networks when facing low-latency M2M applications.", "sentence2": "we show that the Shannon capacity model significantly overestimates the delay performance for such applications, which would lead to insufficient resource allocations.", "label": "entailment", "domain": "Networks"}
{"id": "test_1293", "sentence1": "Based on this novel approach, we provide a performance model for wireless systems that operate at finite blocklength.", "sentence2": "the core contributions of this paper are: We derive probabilistic delay bounds for wireless sys­tems that use channel coding at finite blocklength.", "label": "entailment", "domain": "Networks"}
{"id": "test_1294", "sentence1": "The reason is that the encoder and the decoder do not hold the same version of the reference frame.", "sentence2": "the en-coder does not perform quantization and only has the lossless version of the video frames, while the decoder only has the distorted version.", "label": "entailment", "domain": "Networks"}
{"id": "test_1295", "sentence1": "Generally speaking, the digital video coding efficiency is quite high, although the B-frames should be disabled in a low-delay setting.", "sentence2": "digital transmission in a varying wireless channel is known to have the thresholding effect and saturation effect [15, 14].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1296", "sentence1": "Stereo video (or 3D video) has attracted increasing interests in recent years, and there is a trend that a large portion of the stereo videos will be captured and consumed by mobile devices [1].", "sentence2": "several virtual reality (VR) head-mounted displays (HMDs), including Oculus Rift [4], Sony Morpheu [5] and Microsoft Hololens [3], will be available as full products in the early 2016. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1297", "sentence1": "In a dramatically varying channel, the sender tends to select a robust modulation scheme and a strong channel code to ensure reliability.", "sentence2": "this significantly reduces the effective bandwidth and as a result the source encoder has to operate at a low target video quality.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1298", "sentence1": "The noise trace is used to test the references implementations to ensure fairness.", "sentence2": "we divide the channel SNR into 2dB bins, and all traced data falling into the same bin is averaged for performance computation.", "label": "entailment", "domain": "Networks"}
{"id": "test_1299", "sentence1": "The digital bit stream must be transmitted without error to guarantee basic visual quality and identical side infor­mation at the transmitter and receiver.", "sentence2": "strong protection should be provided to digital transmission.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1300", "sentence1": "In a bigger area with the same MC the distances are longer thus the MC does not have enough speed and sufficient time to keep a high % of overall alive nodes.", "sentence2": "as the network scales the use of local distributed network information achieves good performance-overhead trade-offs.", "label": "neutral", "domain": "Networks"}
{"id": "test_1301", "sentence1": "Considering the rela­tively high performance, we used for our strategies evaluation the exponents in a uniform manner (i.e. all exponents equal to 1).", "sentence2": "according to this strategy the MC is moving along the one dimension of O until it reaches its border.", "label": "neutral", "domain": "Networks"}
{"id": "test_1302", "sentence1": "Conserving the buffer between consecutive optimization windows is particularly useful when the content duration is longer than the optimization window and it is thus not possible to guarantee the QoS over its whole duration. ", "sentence2": "the streaming requests that cannot be scheduled with guaranteed quality must wait for the system to have enough resources for them to start streaming.", "label": "neutral", "domain": "Networks"}
{"id": "test_1303", "sentence1": "To measure the accuracy of COExiST we carry a series of experiments using three nodes, with two nodes represent­ing the secondary network and the third the primary user.", "sentence2": "at the routing layer we use OLSR with an ETX implementation.", "label": "neutral", "domain": "Networks"}
{"id": "test_1304", "sentence1": "However, this strategy only works when the players can correctly determine who is the first player, which may not be possible in distributed applications.", "sentence2": "a player-independent strategy is desired.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1305", "sentence1": "A de­tailed mathematical formulation is derived to estimate the degree of overlap in terms of expected quorum overlap size.", "sentence2": "rDV on a channel is rather influenced by channel quality or CR user s preference on which it wants to achieve RDV on available channels.", "label": "neutral", "domain": "Networks"}
{"id": "test_1306", "sentence1": "According to the definition of ATTR, (presented in section ??) it can only be minimized by increasing the number of RDV which is also illustrates in fig. 8b.", "sentence2": "rDV only achieved if both CR users observe same set of channels.", "label": "neutral", "domain": "Networks"}
{"id": "test_1307", "sentence1": "In asynchronous approaches, the nodes duty cycle schedules are decoupled.", "sentence2": "at an arbitrary instant, only a subset of the nodes are in active mode, providing network services.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1308", "sentence1": "In synchronized approaches, nodes negotiate a schedule in order to align their active or sleep periods.", "sentence2": "the sender node knows when its neighbors are awake and able to receive data packets.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1309", "sentence1": "The selected candidates forward the data packets toward the destination in a prior­itized way, such that a node transmits the packet only if all the high priority candidates failed to do so.", "sentence2": "a packet should be retransmitted only if none of the candidates receives it, which helps to reduce the energy consumption and packet collision.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1310", "sentence1": "Memory management : release buffer space until enough vacant memory space in its generation can store the message.", "sentence2": "there­fore, the concept of Adaptive Collaborative Network Coding of Intra/Inter-Flows based on Bayesian Network(ACNC) is proposed.", "label": "neutral", "domain": "Networks"}
{"id": "test_1311", "sentence1": "However, in a network with multiple generations and various lengths of messages, this scheme would cause less satisfactory performance because it cannot re-adjust the degree dynamically.", "sentence2": "the adaptive collaborative network coding of intra/inter flows(ACNC) method is put forward in section 4 to solve this problem in a further step.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1312", "sentence1": "That is, existing broadcast strategies for DTNs do not perform well with human body mobility.", "sentence2": "however the relative typically small WBAN dimension limits the problem of the flooding algorithms that, with very large networks, makes the retransmissions uncontrollable.", "label": "neutral", "domain": "Networks"}
{"id": "test_1313", "sentence1": "Besides the chosen strategies, these contributions confirmed that the transmissions efficiency increases with the knowledge of the mobility pattern.", "sentence2": "knowing how nodes move allows to reduce the number of unnecessary transmissions and the delivery delay.", "label": "entailment", "domain": "Networks"}
{"id": "test_1314", "sentence1": "However, the discussion presented throughout this paper shows that this approach does not offer any advantage over the plain application of the PWP model.", "sentence2": "the former so­lution makes the modeling task considerably more involved.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1315", "sentence1": "While the quantization error of the local station may be alleviated by newer WiFi chipsets operating on wideband channels (such as the 160 MHz clock of 802.11ac), the vari­ability of the target offset implies that some level of inter­vention of the target station would be needed to increase the precision of the measurement, such as the 802.11v amend­ment which would allow location-related timing information to be shared between local and target stations.", "sentence2": "as of today, there is very limited support of 802.11v features in chipsets and drivers.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1316", "sentence1": "We study the statistical error distribution of the ToF and characterize how it affects the ranging accuracy and precision.", "sentence2": "we aim to answer the following key questions related to timing information extracted for 802.11 ranging: How deterministic and predictable are the time offsets for ranging measurements using regular IEEE 802.11 chipsets?", "label": "entailment", "domain": "Networks"}
{"id": "test_1317", "sentence1": "Sleep Capability Some approaches in the literature consider that the SUs can switch between sleep and awake states to reduce energy consumption during idle periods [6].", "sentence2": "a SU can choose to turn their transceiver off when a PU is detected present, because no secondary transmissions are allowed during that period.", "label": "entailment", "domain": "Networks"}
{"id": "test_1318", "sentence1": "During no change (baseline) or a temporal change the GeMREM sensor sends either a keep-alive signal or just the temporal feature, respectively at a low data rate of 1.16 kbps.", "sentence2": "when there is a change in morphology the GeMREM sensor sends the full data at high data rate (82 kbps).", "label": "contrasting", "domain": "Networks"}
{"id": "test_1319", "sentence1": "In CCN, hierarchically structured names are used to simplify the name space management and aggregation similar to IP.", "sentence2": "iP addresses reveal the physical domain relationships while the hierarchical names cannot.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1320", "sentence1": "In a first evaluation step, we can use this analogy to theoretically estimate the flooding overhead inherited from the topology prior to adaptive FIB optimisations.", "sentence2": "we can derive the worst, best, and average case scenarios for flooding branches of subtrees.", "label": "entailment", "domain": "Networks"}
{"id": "test_1321", "sentence1": "The last class of approaches that we discuss in this paper are countermeasures that analyze PIT consumption per name prefix.", "sentence2": "this observation can be easily misused by an attacker when launching real-world attacks.", "label": "neutral", "domain": "Networks"}
{"id": "test_1322", "sentence1": "The easiest implementation is the request of non-existing content as entries need to expire until they are removed.", "sentence2": "even the request of existing content may harm the infrastructure when the entries in the Pending Interest Table (PIT) exceed the content delivery rate (e.g., due to large network delays) [5].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1323", "sentence1": "The Pi boxes scan for bluetooth devices and keep track of the devices that they see.", "sentence2": "the multicast features of CCN allows to forward an alert to multiple end points: the gateway of the function, but also the tag's user or the local authority to retrieve the missing tagged item.", "label": "neutral", "domain": "Networks"}
{"id": "test_1324", "sentence1": "Such a service is not valuable until it has many users to detect tags.", "sentence2": "since early adopters would only see little value until that critical mass is reached, they would be less likely to encourage others to adopt.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1325", "sentence1": "One popular technique for flat name is Distributed Hashing Table (DHT) based approach [3-5], where multiple servers form circular linked list and the bindings are stored in the appropriate server.", "sentence2": "we will demonstrate the benefits of utilizing BFs by our prototype.", "label": "neutral", "domain": "Networks"}
{"id": "test_1326", "sentence1": "The current Internet is addressing on the order of 109 nodes, whereas the number of addressable ICN objects is expected to be several orders of magnitude higher.", "sentence2": "scalability on the ever-increasing number of NDO becomes one of the most important challenges on designing NRS.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1327", "sentence1": "Initially, dashboards will be statically defined.", "sentence2": "we encourage researchers to contribute reports at the completion of research projects, and moving forward we hope to allow everyone to dynamically define new reports.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1328", "sentence1": "We will continue to maintain the scans.io interface, provide continued access to our historical datasets, and allow researchers to upload other data.", "sentence2": "we will no longer post our regular scans to https://scans.io, but rather encourage users to download these directly from Censys's web interface.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1329", "sentence1": "Finally, it is our hope that by publishing scan data, carefully acquired and properly curated, we can reduce the need for Internet scanning performed by other researchers, and thus reduce the overall burden on destination networks.", "sentence2": "it is well established that attackers already use Internet-wide scanning to find vulnerable machines from botnets and bullet-proof hosting providers [17].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1330", "sentence1": "Ac­cording to Shannon Theorem [1], a large SNR can support a high speed service than a small SNR on the same channel bandwidth.", "sentence2": "we would like to enable a receiver at the desired location to always achieve a large SNR, and an eavesdropper at an undesired location to encounter a low SNR, so that it cannot distinguish the received signal from the background noise and fails to decode received data.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1331", "sentence1": "We observe a clear and fluent video when the receiver is located at the desired position, and the video quality deteriorates when the receiver moves away from undesired locations.", "sentence2": "we encounter frequent video stuck while playing, and severely distorted images.", "label": "entailment", "domain": "Networks"}
{"id": "test_1332", "sentence1": "Service providers may use existing localization algorithms like time-ofarrival (TOA) and angle-of-arrival (AOA) to find the locations of wireless users, and encrypt the service data so that users at target locations can use appropriate keys to decrypt it.", "sentence2": "cryptographic encryption may cause a significant latency, and thus fail to support common services like high-speed downloading and online video watching.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1333", "sentence1": "At other locations where constructive in­terference vanishes, j1 and j2 do not cancel each other, and instead they serve as jamming signals to decrease the SNR at receivers at these locations.", "sentence2": "the receivers will experience a service of bad quality.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1334", "sentence1": "Note that the propagation synchronization may introduce overlapping time slots due to the varying time shifts experienced by different users.", "sentence2": "proper time guard should be inserted between time slots to eliminate the overlaps and avoid the interference among multiple users.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1335", "sentence1": "Assume an ideal synchro­nization algorithm is in use and these packets arrive at the receiver at the service location simultaneously.", "sentence2": "they constructively interfere with each other to form a boosted received packet whose magnitude is twice of that of an indi­vidual packet.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1336", "sentence1": "Effect Multipath effect is the phenomena that signals sent by the transmitter travel along multiple paths to reach the receiver.", "sentence2": "since users are located at different locations, the transmit signals may need to be sent at distinct times to compensate the time difference, and thus asynchronous CDMA scheme is required at the transmitter.", "label": "neutral", "domain": "Networks"}
{"id": "test_1337", "sentence1": "Also, a malicious FWD could collaborate with a malicious IdP and send information about the RP to the IdP, and hence, undermine privacy.", "sentence2": "for our system to provide authentication and privacy, we require that FWDs behave honestly.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1338", "sentence1": "For the proof, we first show some general properties of SWS auth.", "sentence2": "we show that encrypted communication over HTTPS between an honest relying party and an honest IdP cannot be altered by the (network) attacker, and, based on that, any honest relying party always retrieves the “correct” public signature verification key from honest IdPs. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1339", "sentence1": "This is the first SSO system which respects user's privacy.", "sentence2": "the system allows users to log in to RPs with their email addresses.", "label": "neutral", "domain": "Networks"}
{"id": "test_1340", "sentence1": "Informally, these properties can be stated as follows: (A) The attacker should not be able to use a service of an honest RP as an honest user.", "sentence2": "the attacker should not get hold of (be able to derive from his current knowledge) a service token issued by an honest RP for an ID of an honest user (browser), even if the browser was closed and then later used by a malicious user, i.e., after a CLOSECORRUPT (see Section 3.3).", "label": "entailment", "domain": "Networks"}
{"id": "test_1341", "sentence1": "RPdoc sends this address in a POST request to RP 3 . ", "sentence2": "the third component, script, is an injective mapping from a script in S to its string representation script(s) (a constant in Σ) so that it can be part of a messages, e.g., an HTTP response.", "label": "neutral", "domain": "Networks"}
{"id": "test_1342", "sentence1": "This full day tutorial on synchronization and security in Named Data Networking (NDN) will share important architectural concepts we are exploring in these areas, the software we have built to perform these tasks, and remaining open issues.", "sentence2": "it will emphasize how the existing open source toolset provides a platform for exploring the open research questions.", "label": "entailment", "domain": "Networks"}
{"id": "test_1343", "sentence1": "This tutorial will share important architectural concepts we are exploring in these areas, the software we have built to perform these tasks, and remaining open issues.", "sentence2": "it will emphasize how the existing open source toolset provides a platform for exploring the open research questions.", "label": "entailment", "domain": "Networks"}
{"id": "test_1344", "sentence1": "Tactical scenarios normally cannot rely on such infrastructure and events like natural disasters can severely damage the network infrastructure in rural and urban environments.", "sentence2": "there is a need to develop solutions that provide SoA-based application and services running on heterogeneous and often constrained devices that compose tactical and mobile ad-hoc networks with Quality of Service (QoS) levels that meet their requirements.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1345", "sentence1": "Network traffic characterization constantly analyzes the volume of traffic in the network to provide the decision making component of NetProxy with updated information about what nodes and applications are generating traffic, what types of data are being transmitted, the current bandwidth consumption, and observed radio/link performance.", "sentence2": "the QoS related decision making process in NetProxy is both application- and network- aware.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1346", "sentence1": "ACM NetProxy [3] [4] provides transparent integration between networked applications and the ACM.", "sentence2": "as shown by experimental results obtained during a test in a field demonstration event, GM enables multiple nodes in subnetworks to benefit from NetProxy and the ACM.", "label": "neutral", "domain": "Networks"}
{"id": "test_1347", "sentence1": "When either happens, NetProxy intercepts the ARP request, changes the source hardware address (SHA) of the requester with that of the ENI of the proxy gateway, and finally forwards the modified packet on the other network interface.", "sentence2": "similarly, Figure 3b shows that the density distribution of the number of packets in the EN during each interval has much gentler slopes than that describing the conditions in the IN in the same intervals.", "label": "neutral", "domain": "Networks"}
{"id": "test_1348", "sentence1": "Consequently, expansions were made to networks of roads to accommodate the increasing number of vehicles.", "sentence2": "these changes resulted in an increase in the rate of deterioration of road health, out-pacing current road health monitoring techniques and increasing the number of accidents.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1349", "sentence1": "The significant increase in the number of drivers can be put to use to help address these challenges, through continuous monitoring and reporting of events by the drivers and their vehicles.", "sentence2": "to enlist the help of the public and their vehicles the system must be inexpensive to attract participants.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1350", "sentence1": "Emergency and periodic messages are transmitted over the CCH and infotainment messages are transmitted over SCH.", "sentence2": "to transmit these messages, the transmitter is switched alternately between CCH and SCH in a 100 ms long Synchronization Interval (SI), as shown in Figure 1.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1351", "sentence1": "As discussed ear­lier, periodic messages are generated after every SI.", "sentence2": "when a safety critical event occurs, an emergency message is generated which attains higher priority than periodic messages for transmission in the CCH [3].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1352", "sentence1": "Last, we evaluate the online detecting lag and the offline training time of Opprentice (§5.8).", "sentence2": "since anomalies occur less frequently in practice, an arbitrary training set is unlikely to cover enough anomalies [16].", "label": "neutral", "domain": "Networks"}
{"id": "test_1353", "sentence1": "The above describes how to configure cThlds based on the PCScore in an offline or “oracle” mode.", "sentence2": "we configure cThlds after the data to be detected (also called a test set) have already arrived. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1354", "sentence1": "For example, stop growing the tree earlier after it exceeds a threshold of depth.", "sentence2": "it is still quite tricky to determine such a threshold.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1355", "sentence1": "LPR reveals the actual usage of MPLS according to the inferred label distribution protocol and is able to make the distinction between ECMP and TE multi-path forwarding.", "sentence2": "the distribution given in Fig. 9 does not show the Mono-LSP case9 . ", "label": "neutral", "domain": "Networks"}
{"id": "test_1356", "sentence1": "This means we do not observe transit tunnel diversity, the same LSP being always used whatever the destination.", "sentence2": "for this tunnel, we are not able to reveal ECMP load balancing (by definition of the class) or the deployment of several FECs used to reach different ASes with different routing constraints (although 4 we consider at least two destination ASes as stated in the filtering subsection).", "label": "reasoning", "domain": "Networks"}
{"id": "test_1357", "sentence1": "In practice, if the vast majority of LSPs disappear for a given AS, we reinject the whole set of its LSPs to perform a standard classification on a given snapshot4 . ", "sentence2": "we do not remove such an AS and continue to process it as the others but adding a dynamic tag to it.", "label": "entailment", "domain": "Networks"}
{"id": "test_1358", "sentence1": "It must be applied once the data has been collected, as long as this data contains information related to MPLS tunnels [7, 8].", "sentence2": "further, we are able to observe the evolution of each Autonomous System (AS) independently and understand whether it enables path diversity, how, and when it evolves (e.g., from almost no path diversity to a wide deployment of TE).", "label": "neutral", "domain": "Networks"}
{"id": "test_1359", "sentence1": "Based on an extensive and longitudinal traceroute dataset obtained from CAIDA, we apply LPR and find that each ISP behavior is really specific in regard to its MPLS usage.", "sentence2": "we are able to ob­serve independently for each ISP the MPLS path diversity and usage, and its evolution over time.", "label": "entailment", "domain": "Networks"}
{"id": "test_1360", "sentence1": "These results suggest that by studying the PLC network and its temporal variation, probing can be optimized to achieve a good tradeoff between overhead and accuracy.", "sentence2": "81% of station pairs that are connected by PLC links, are also connected by WiFi links.", "label": "neutral", "domain": "Networks"}
{"id": "test_1361", "sentence1": "It turns out that the devices maintain the channel-estimation statistics, as the estimated capacity resumes from the previous value before stopping the probing process.", "sentence2": "the convergence time of the capacity estimation does not apply in realistic probing conditions.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1362", "sentence1": "It turns out that the devices maintain the channel-estimation statistics, as the estimated capacity resumes from the previous value before stopping the probing process.", "sentence2": "the variation of µ is governed by the electrical load.", "label": "neutral", "domain": "Networks"}
{"id": "test_1363", "sentence1": "It also enables a sniffer mode with which we can capture the SoF delimiters of all received PLC frames.", "sentence2": "hence, we create two different networks, shown with different colors in Figure 2.", "label": "neutral", "domain": "Networks"}
{"id": "test_1364", "sentence1": "100% of station pairs that are connected with WiFi are also connected with PLC.", "sentence2": "81% of station pairs that are connected by PLC links, are also connected by WiFi links.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1365", "sentence1": "The authors find that hybrid networks contribute to increase coverage in home networks; they also argue that using alternating technologies for multi-hop routes yields good performance.", "sentence2": "they do not study link metrics that can be used to optimize routing in such networks.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1366", "sentence1": "It shows that P Ber r decreases as throughput increases, as expected.", "sentence2": "because the tone maps are updated based on this metric, some average links might have lower P Ber r than the best links of the testbed.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1367", "sentence1": "Along these same lines, we argue that the periodicity present in Figure 3 stems from users sending less business mail on week ends and instead relying on personal accounts provided by major providers.", "sentence2": "on weekdays between April 1 and 26, 2015, Gmail encrypted 79.8% of outbound messages, while mail servers encrypted 53.7% of incoming connections.", "label": "entailment", "domain": "Networks"}
{"id": "test_1368", "sentence1": "In terms of sheer volume, during April 2015, Gmail was able to validate 94% of inbound messages using a combination of DKIM (83%) and SPF (92%).", "sentence2": "among the Alexa Top Million mail servers, only 47% deploy SPF policies and only 1% provide a DMARC policy, the absence of which leaves recipients unsure whether an unsigned message is invalid or expected.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1369", "sentence1": "In our scan, 414,374 Top Million domains (52% of domains with valid SMTP servers, and 64% of domains that support START-TLS) present certificates that validate against the Mozilla NSS root store [38], as detailed in Table 5.", "sentence2": "only 0.6% of domains present trusted certificates that match their domain name, while 34.2% present trusted certificates that match their MX server.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1370", "sentence1": "We identify most parking domains with more than one of our three methods.", "sentence2": "we identify all but 124 of nearly 280,000 domains on our set of parking name servers with another approach.", "label": "entailment", "domain": "Networks"}
{"id": "test_1371", "sentence1": "Defensive registrants purchase domains to defend the string but with no intent to develop content, while speculative registrants purchase domains to resell later with no intent to develop content.", "sentence2": "speculative registrants are monetarily motivated on a per-domain margin, while defensive registrants have revenues outside the domain business.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1372", "sentence1": "To map Web pages to inputs for a clustering algorithm, we follow a conventional “bag-of-words” approach which extracts HTML features from the Web pages. ", "sentence2": "we compose a dictionary of all terms that appear in the HTML source code, and for each Web page, we count the number of times that each term appears. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1373", "sentence1": "ICANN intends the use of WHOIS for any lawful purpose except to enable marketing or spam, or to enable high volume, automated processes to query a registrar or registry's systems [14].", "sentence2": "iCANN encourages its use by consumers, registrars, and law enforcement, and discourages its use by spammers [29].", "label": "entailment", "domain": "Networks"}
{"id": "test_1374", "sentence1": "Thus, we iterated this approach to achieve greater coverage.", "sentence2": "we clustered the remaining unlabeled Web pages, manually inspected and labeled homogenous clusters, and performed thresholded nearest neighbor classification now with a larger set of labeled examples.", "label": "entailment", "domain": "Networks"}
{"id": "test_1375", "sentence1": "Our analysis shows that smartphone users select appropriate network interfaces taking into account the deployment of emerging technologies, their bandwidth demand, and their economic constraints.", "sentence2": "users show diversity in both how much traffic they send, as well as on what networks they send it.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1376", "sentence1": "In our measurements, the median cellular download traffic volume is 36MB/day and the median WiFi download traffic is 51MB/day in 2015 (§ 3.2).", "sentence2": "58% of smartphone traffic is WiFi; a 1.4:1 ratio of WiFi-to-cellular traffic.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1377", "sentence1": "In particular, in WiFi networks at home, this category accounts for a high percentage of traffic volume due to online storage software that uploads/downloads large files only if a WiFi interface is available.", "sentence2": "applications seem to play a major role in promoting WiFi offloading.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1378", "sentence1": "Our results show slow but clear growth of WiFi traffic offloading during these three years.", "sentence2": "the deploy­ment of public WiFi networks provides users both simple network connectivity and also more bandwidth for bandwidth-­intensive applications such as video streaming and software updates.", "label": "entailment", "domain": "Networks"}
{"id": "test_1379", "sentence1": "Most commuters in this area use public transportation (e.g., trains, subways, and buses) rather than personal cars.", "sentence2": "the probability to encounter public WiFi networks is likely high, and resulting WiFi traffic volume in public spaces is also high.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1380", "sentence1": "The success of these WiFi deployment hinges on this key question: how do smartphone users select a network from the alternatives available to them?", "sentence2": "wiFi-user ratio differs between the two device OSes rather than cellular carriers.", "label": "neutral", "domain": "Networks"}
{"id": "test_1381", "sentence1": "The percentage of the productivity category increases in upload volume.", "sentence2": "in WiFi networks at home, this category accounts for a high percentage of traffic volume due to online storage software that uploads/downloads large files only if a WiFi interface is available.", "label": "entailment", "domain": "Networks"}
{"id": "test_1382", "sentence1": "The application usage and resulting performance in cellular net­works also indicate large diversity [47, 16, 28, 51, 44, 27].", "sentence2": "the usage pattern of applications in smart­phones depends on the mobility and geographical region of users [47, 51, 52].", "label": "entailment", "domain": "Networks"}
{"id": "test_1383", "sentence1": "For iOS devices, the software does not report detailed in­formation about the WiFi interface.", "sentence2": "we conclude that iOS devices connect to WiFi 30% more than do Android devices, as shown in Figure 9(c).", "label": "contrasting", "domain": "Networks"}
{"id": "test_1384", "sentence1": "Accurate Detection: The main idea underlying our diagnostic application here is as follows: Elements in the virtualization stack deliver packets to each other via intermediate buffers or function calls, and they typically use nonblocking I/O in doing so.", "sentence2": "if an element cannot write to its successor, or its target buffer is full, packets get dropped.", "label": "entailment", "domain": "Networks"}
{"id": "test_1385", "sentence1": " Likewise, in Figure 5, the shared datapath of input traffic and output traffic (purple/solid arrows) can be the software resource in contention; similar contention can arise for other software resources such as shared buffers/queues.", "sentence2": "the remaining middleboxes in the candidate set are returned (line 18) as the plausible root cause middleboxes.", "label": "neutral", "domain": "Networks"}
{"id": "test_1386", "sentence1": "Whereas traditional dataplanes consisted just of hardware switching elements and network links connecting network end points, the advent of NFV means that we need to rethink what constitutes the data plane.", "sentence2": "it now also includes the software components shown above that are traversed within middlebox VMs.", "label": "entailment", "domain": "Networks"}
{"id": "test_1387", "sentence1": "At this point, using PerfSight the operator identifies that the TUN of load balancer 2 is dropping packets and it is in an Overloaded state.", "sentence2": "the operator has identified tenant 2 s bottleneck.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1388", "sentence1": "An AS will prefer routes through a neighboring customer, then routes through a neighboring peer, and then routes through a provider.", "sentence2": "an AS will prefer cheaper routes.", "label": "entailment", "domain": "Networks"}
{"id": "test_1389", "sentence1": "Those that are actually connected experience very poor performance.", "sentence2": "de Telecomunicaciones de Cuba, S.A. (ETECSA) to provide direct international long distance telephony between the USA and Cuba [16].", "label": "neutral", "domain": "Networks"}
{"id": "test_1390", "sentence1": "The Internet suffers from well-known performance, reliability, and security problems.", "sentence2": "proposed improvements have seen little adoption due to the difficulties of Internet-wide deployment.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1391", "sentence1": "In this case, the camera can perfectly capture the displayed image.", "sentence2": "if the dis­played image is a visual code of any kind, all data from the code can be extracted.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1392", "sentence1": "In our demo, we show how our Focus prototype copes with different channel impairments.", "sentence2": "we will showcase Focus’s performance on smartphones with different capabilities, ranging from older models with poor cameras to recent smartphones with high-resolution cameras and high-quality lenses.", "label": "entailment", "domain": "Networks"}
{"id": "test_1393", "sentence1": "Al­though RPL decouples the routing metric from the main standard, the most widely used routing metric is the com­bination of ETX and hop distance.", "sentence2": "this routing metric cannot be used in RPL over BLE since BLE MAC (in Figure 1) does not provide the number of retransmissions for the upper layers.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1394", "sentence1": "This is because developers do not need to choose a suitable protocol out of many existing protocols.", "sentence2": "in the first experiment, the 94 nodes generate packets and we vary IPI (Inter-Packet Interval).", "label": "neutral", "domain": "Networks"}
{"id": "test_1395", "sentence1": "In a full-duplex wireless network, an access point and sensor nodes transmit information and power signals simul­taneously.", "sentence2": "this simultaneous transmission leads to a collision problem.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1396", "sentence1": "Third, data should be downloaded from the nodes in a fair way.", "sentence2": "the amount of data collected from each node should be greater than a certain application-specific threshold.", "label": "entailment", "domain": "Networks"}
{"id": "test_1397", "sentence1": "Based on the two experiments discussed in the previous section, we observe that the WPT efficiency is one of key factors for the scheduling model.", "sentence2": "fS-WPT is proposed to prioritise the nodes for scheduling based on a ratio of the link quality and harvested energy.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1398", "sentence1": "Although Broadcast Encryption (BE) [3] exhibits excellent distinctions, most BEs are k-resiliency systems, where k + 1 exposed keys can collude to generate shared secrets.", "sentence2": "although our storage overhead is linear, it is feasible in practical settings as it is quantified in Fig. 1(c).", "label": "neutral", "domain": "Networks"}
{"id": "test_1399", "sentence1": "There have been many efforts geared towards improving energy efficiency of different components of the system including data sampling, query processing, radio communication and processor duty cycling.", "sentence2": "these individual components may not be flexible to adapt to changing network and application dynamics.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1400", "sentence1": "After a reboot, anti-replay data stored in RAM get lost.", "sentence2": "to prevent replay attacks after re­boots, anti-replay data must be stored across reboots.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1401", "sentence1": "The uncontrolled movement experiment is carried out in the same environment as the controlled movement experiment.", "sentence2": "the transmitter and the 4 receivers are located in different rooms as shown in Figure 10.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1402", "sentence1": "The estimation result -the CSI -is used by a receiver to extract the transmitted information.", "sentence2": "as the CSI depends on the communication environment and the transmitter hardware, it can be used as well for security purposes.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1403", "sentence1": "By different defected ground structures (DGS) like folded slot, dumbbell shapes slot, interdigital slot, meander slot are used for realize dual passband filters [3-6].", "sentence2": "these bands utilized for Satellite communication, mobile communication and Aeronautical radio navigation.", "label": "neutral", "domain": "Networks"}
{"id": "test_1404", "sentence1": "Cloud Computing is the latest technology that has come to the center-stage because of its ability to achieve economies of scale, quite like mass-production.", "sentence2": "shifting to Cloud comes with certain strings attached, especially security concerns associated with outsourcing critical data and processing to a third party.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1405", "sentence1": "And when we use the new approach only 8 users are connected to the MBS.", "sentence2": "the new algorithm has reduced the load on MBS which is our main objective.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1406", "sentence1": "Also this algorithm is developed for only one macro base station.", "sentence2": "its feasibility for more than one macro base station also needs to be checked.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1407", "sentence1": "If we can utilize ubiquitous WiFi devices, it will be a major breakthrough in terms of costs, feasibility, universality, etc..", "sentence2": "wiFi signals suffer from limited bandwidth and insufficient time resolution, which brings challenges for WiFi based detection [3] [4].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1408", "sentence1": "We detail our two human detection schemes in section 4 and 5, respectively.", "sentence2": "context-aware technology is of rising demand in our everyday life, which aims to sense the variations of surrounding environment through deployed sensors or related technologies.", "label": "neutral", "domain": "Networks"}
{"id": "test_1409", "sentence1": "Managing remotely deployed sensors has been thoroughly studied in the context of WSNs.", "sentence2": "existing WSNs solutions exhibit poor efficiency due to a wide variety of emerging IoT issues such as mobility, heterogeneity, scale, connectivity, security and privacy, energy, and the ease of management.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1410", "sentence1": "The structured overlays have been widely used in a variety of distributed systems.", "sentence2": "the structures of smallworld networks (originated in [17]) inspired several popular DHT designs, e.g., Chord [28] and Symphony [20].", "label": "entailment", "domain": "Networks"}
{"id": "test_1411", "sentence1": "This work was supported by the Director, Office of Sci­ence, Office of Advanced Scientific Computing Research (ASCR), of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231.", "sentence2": "tCP interprets packet loss as network congestion, and reduces its sending rate when loss is detected.", "label": "neutral", "domain": "Networks"}
{"id": "test_1412", "sentence1": "From these results we are comfortable making the claim that pacing flows does yield the effects that our daemon looks for when selecting candidates for pacing. ", "sentence2": "our daemon would not select this same flow for pacing again, given the fact that the retransmissions have fallen to 0.", "label": "entailment", "domain": "Networks"}
{"id": "test_1413", "sentence1": "However, the very mechanisms that make TCP so survivable also make it perform poorly when network conditions are not ideal. ", "sentence2": "tCP interprets packet loss as network congestion, and reduces its sending rate when loss is detected.", "label": "entailment", "domain": "Networks"}
{"id": "test_1414", "sentence1": "After developing and deploying our traffic shaping tool on the ESnet testbeds, we modified the tool so that it would only run in a monitor only capacity.", "sentence2": "it would not be able to actually shape any traffic, only monitor traffic from the user-space.", "label": "entailment", "domain": "Networks"}
{"id": "test_1415", "sentence1": "What is immediately visible, is that the download speeds are very variable.", "sentence2": "for the green node (the one with rel­atively stable RTT times), most measurements lie around the same average throughout the duration of the tests.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1416", "sentence1": "Pro­posals such as nutrition labels [24] could also help empower consumers by making them more aware of the QoS to expect and then use the information to make better choices when buying data plans.", "sentence2": "unless TRAI does not mandate some minimum QoS standards to which providers can be held accountable, or the published information is not made available to consumers easily to be able to exercise their choice in selecting providers, even these stronger regu­latory measures may arguably not yield much benefits.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1417", "sentence1": "In addition, it is not allowed to use RC-QP for all source-destination pairs because RC-QPs consume prohibitive amount of memory with millions of MPI processes.", "sentence2": "the MPI library need to select mapping of RC-QPs and DC-QPs according to the current communication pattern.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1418", "sentence1": "DC, on the other hand, consumes less memory than RC but its performance drops when sending messages to different destinations or when many DCs sends a message to the same destination DC.", "sentence2": "the library should find the best mapping of RCs and DCs to pairs of source peer and destination peer according to the communication pattern of the application.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1419", "sentence1": "They conjectured that this was due to network contention, but did not investigate it via lower-level analysis.", "sentence2": "the performance predictability of applications over moderate time periods could be pre­dicted given a model of application performance to network contention, or given a sufficient amount of empirical data, could be predicted based on past application performance with the given network environment.", "label": "neutral", "domain": "Networks"}
{"id": "test_1420", "sentence1": "Besides, this area has a complex ocean tidal current due to Kuroshio Current and local currents, which results in the most complicated flow distribution area along China's sea-coast.", "sentence2": "zRS Island is a suitable place for tide and current observation.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1421", "sentence1": "Usually, expedition ship are armed with various observatory equipment and massive personnel, and can only do the observatory along its route. ", "sentence2": "along with expedition ship, the moorings system are also developed.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1422", "sentence1": "We assume that a beacon AUV is able to obtain its accurate absolute location via satellite.", "sentence2": "there is no mean error for localization of beacon AUVs.", "label": "entailment", "domain": "Networks"}
{"id": "test_1423", "sentence1": "According to previous studies [12-14], prolonging of the acoustic transmission range leads to narrower bandwidth, lower communication reliability, worse localization accuracy and more energy consumption.", "sentence2": "a multi-hop communication pattern is more practical for MUWNs [15, 16].", "label": "reasoning", "domain": "Networks"}
{"id": "test_1424", "sentence1": "Unfortunately, in SpaceHub, we cannot assume that the signal structure is always available (i.e., being wireless technology agnostic).", "sentence2": "we cannot acquire channel coefficients through such measurements.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1425", "sentence1": "For example, Wi-Fi, ZigBee, Bluetooth, Digital cordless phone, and many other proprietary wireless technologies for surveillance camera or remote controller, all operate on the same 2.4GHz ISM band.", "sentence2": "interference among these wireless technologies may result in unreliable com­munication or low network throughput.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1426", "sentence1": "The core idea is that all separated signal copies from one unique source should show a strong correlation, whereas signal streams from different sources are uncorrelated.", "sentence2": "spaceHub clusters all the separated signal streams into several groups according to their cross correlation.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1427", "sentence1": "It is not straightforward as the signal structure is not available, and therefore we will not be able to identify the destination by decoding the MAC address.", "sentence2": "similar to the identification of multipath signals, SpaceHub applies a preprocessing algorithm to cluster signal copies based on cross correlation results.", "label": "neutral", "domain": "Networks"}
{"id": "test_1428", "sentence1": "Path-end validation does not protect against the \"2-hop attack\", in which AS 2 pretends to be directly connected to a (legacy) neighbor of AS 1 (say, announces the bogus route 2 - 40 - 1).", "sentence2": "such attacks turn out to be quite ineffective, since the path that the attacker can announce to the victim's prefix, must consist of at least two hops, and BGP paths are typically short (about 4-hops-long on average [24]), i.e., most ASes will not fall victim to the attack.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1429", "sentence1": "Allowing the original scheduling algorithms to be preemptive allows packets to be fragmented, which then makes replay extremely difficult even in simple networks (with store-and-forward routers).", "sentence2": "disallowing preemption in the candidate UPS overly limits the flexibility and would again make replay impossible even in simple networks. ", "label": "contrasting", "domain": "Networks"}
{"id": "test_1430", "sentence1": " Furthermore, the small increased memory consumption of UNIVMON comes at a dramatically increased flexibility and generality across the suite of applications. ", "sentence2": "openSketch is in effect using K-times as many resources if there are K distinct tasks as it instantiates a separate sketch when we look at the set of applications, whereas UNIVMON uses a single universal sketch instance.", "label": "entailment", "domain": "Networks"}
{"id": "test_1431", "sentence1": "We assume that we have a single switch processing this trace and that in both cases the “controller” periodically polls the switch for the sketch every 5 seconds. ", "sentence2": "the memory numbers reported are roughly for a 5-second trace. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1432", "sentence1": "For brevity, we focus on each of these metrics computed over one feature, namely the source IP address.", "sentence2": "we know G is in Stream-PolyLog and we approximate G-sum in polylogarithmic space using the uni­versal sketch.", "label": "neutral", "domain": "Networks"}
{"id": "test_1433", "sentence1": "Sekar et al, showed empirically that flow sampling and sample-and-hold [24] can provide comparable accuracy to sketching when equipped with similar resources.", "sentence2": "the proof of this theorem is outside the scope of this paper and we refer readers to the previous work of Braverman et al [9].", "label": "neutral", "domain": "Networks"}
{"id": "test_1434", "sentence1": "A recent proposal uses secure multi-party computation (SMPC) to achieve policy privacy [17].", "sentence2": "transforming ar­ bitrary computation to SMPC is non-trivial, and it suffers from severe performance issues.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1435", "sentence1": "Many new designs for inter-domain routing aim at providing new features (i.e., flexible route selection [16, 36, 37] and verifiability [26, 39]).", "sentence2": " in inter-domain routing, introducing new features often comes at the cost of privacy, which is something that ISPs are not willing to sacrifice [17, 39].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1436", "sentence1": "Existing approaches, such as SPIDeR [39], propose to verify whether ASes live up to their promises by collaboratively verifying the promise.", "sentence2": "sPIDeR relies upon the condition that all neighbors of a promise-breaking AS correctly follows their design to perform the verification.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1437", "sentence1": "Motivated by the recent movement towards commoditization of trusted exe­cution environments (TEEs), this paper explores alternative design choices that application and protocol designers should consider.", "sentence2": "we explore the possibility of using Intel SGX to provide security and privacy in a wide range of network applications.", "label": "entailment", "domain": "Networks"}
{"id": "test_1438", "sentence1": "We believe many research opportunities remain in realizing or extending our suggested models.", "sentence2": "one notable fact is that unlike other TEEs, its trusted computing base (TCB) only includes the processor and the application code inside the enclave.", "label": "neutral", "domain": "Networks"}
{"id": "test_1439", "sentence1": "\u0018is is because updated baselines’ extra control information cannot be disseminated across BGP gulfs.", "sentence2": "we end up with this requirement: UB-R1 Disseminate updated baseline’s additional control information across gulfs.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1440", "sentence1": "An AS that supports MIRO [24] offers alternate paths for payment (the rightmost one).", "sentence2": "the transit AS cannot discover the MIRO-enabled AS because BGP does not allow discovery of ASes custom services or the extra coordination required to use them.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1441", "sentence1": "The second feature, adding value to other protocols, is also consistent with the notion that network control functions that benefit other protocols should have an easier time be­ing adopted.", "sentence2": "its high p-value together with the small number of protocols in this category, make it difficult to draw strong conclusions from its selection.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1442", "sentence1": "To meet the requirement II, all devices including non­mesh devices and mesh APs must be authenticated to join the WMN by EAP authentication method such as PEAP or EAP-TLS.", "sentence2": "a mesh AP need to be an authenticator as an entry point to the WMN and also a supplicant connecting another mesh AP.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1443", "sentence1": "As for the fair comparison, poptrie should be compared to LPMnoSSE result, since both does not employ SSE optimization.", "sentence2": "poptrie may benefit from SSE optimization to obtain a better performance in the future.", "label": "entailment", "domain": "Networks"}
{"id": "test_1444", "sentence1": "We then consider all pairs (ti, tj ) i = j, and consider a discretised grid counting the number of services in buckets of size 10.", "sentence2": "we count the number of pairs falling in each bucket.", "label": "entailment", "domain": "Networks"}
{"id": "test_1445", "sentence1": " Indeed, CDFs now saturate at Sim = 0.3, meaning that the activity of different users is in 70% of cases very different.", "sentence2": "people keep accessing previously unseen services, so that the total number of distinct services keeps growing over time. ", "label": "neutral", "domain": "Networks"}
{"id": "test_1446", "sentence1": "As shown, the simi­larity among different users is very small, and smaller than the one among the same user, cfr.", "sentence2": "intuitively, we want to check how many times the most similar user turns out to be u in a group of a 1000 users {ui}.", "label": "neutral", "domain": "Networks"}
{"id": "test_1447", "sentence1": "This means that, independently of the volume of their activity, the sets of services contained in time bins are significantly different.", "sentence2": "users tend to contact same services over time, but the number of new services is however fairly large, meaning that the degree of repetitiveness is low.", "label": "entailment", "domain": "Networks"}
{"id": "test_1448", "sentence1": "In this work, we focus on the exploration of users habits when browsing the Web.", "sentence2": "we take the point of view of the network, from where DNS re­quests of users are observable, and from which we can extract the names of the services they are accessing.", "label": "entailment", "domain": "Networks"}
{"id": "test_1449", "sentence1": "We are currently designing an experiment aimed at overcoming the threats to conclusion validity.", "sentence2": "it is based on a model capable of simulating parallel communication attempts, and that implements the timeout mechanism so that the effects of the transmission delay can be taken into account.", "label": "entailment", "domain": "Networks"}
{"id": "test_1450", "sentence1": "Therefore, future work should replicate the experiment here con-ducted in order to increase confidence in the results.", "sentence2": "the design of the replicated experiments must face the remarked threats to validity and pitfalls.", "label": "entailment", "domain": "Networks"}
{"id": "test_1451", "sentence1": "The size of each routing table changes as a result of both the discovery of a new route (so a new entry is added in the routing table), and a link breakage (so an existing entry is removed from the routing table).", "sentence2": "the size of each routing table can increase or decrease during the simulation execution.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1452", "sentence1": "In the latter, the aggregation process is monopolized by access points.", "sentence2": "we actually define individual preference relationships us­ing utility functions for B-APs and P-APs.", "label": "neutral", "domain": "Networks"}
{"id": "test_1453", "sentence1": "The closest clusters to the BS have smaller sizes compared to those farther away.", "sentence2": "the amount of intracluster traffic is significantly reduced and the nearest sensor nodes to the BS consume less energy than distant sensor nodes.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1454", "sentence1": "In Section 5 the performance of the proposed approach is evaluated.", "sentence2": "if the APs are as­signed non-bonded channels only then n clients may suffer by operating with a reduced transmission rate and thereby degrading the overall network performance.", "label": "neutral", "domain": "Networks"}
{"id": "test_1455", "sentence1": "The positive impact of using a bonded channel is that the n clients can achieve a maximum data rate of 300 Mbps from their associating AP and thereby im­proving the network throughput [2].", "sentence2": "use of bonded channel may increase co-channel interference among the APs and thus degrading the network performance.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1456", "sentence1": "Typically, the payload size is known in advance for this type of traffic.", "sentence2": "the absence of periodicity makes it difficult to calculate the bandwidth requirement.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1457", "sentence1": "Energy conservation, ambient energy harvesting, incremental deployment, battery replacement, and wireless charger are among the typical suggested solutions.", "sentence2": "the majority of these solutions didn't revisit the standard specification to implement the new proposed techniques.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1458", "sentence1": " The connected M2M nodes will include both resourceconstrained and resource-rich devices.", "sentence2": "their capabilities of computation power and memory size are heterogeneous.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1459", "sentence1": "InitCond: This condition requires that any initial state returned  by the implementation's init function has a 0 received-message count according to Rec.", "sentence2": "the initial state should not have received any update from any node.", "label": "entailment", "domain": "Networks"}
{"id": "test_1460", "sentence1": "In a DDoS attack, an attacker may spoof requests from UACs in order to cause the SIP proxy server to allocate resources to processing the spoofed requests, thus starving out legitimate requests (denial of service).", "sentence2": "an attacker sends an INVITE request, but with a spoofed address from a UAC pool.", "label": "entailment", "domain": "Networks"}
{"id": "test_1461", "sentence1": "The virtualization of IP telecommunications networks provides an opportunity for deeper knowledge about security to be embedded inside such networks through a combination of real-time analysis of network behavior and the leveraging of the dynamic reconfiguration capabilities of virtualized networks.", "sentence2": "this combination of analytics and autonomics offers the potential for networks to automatically detect security threats in real-time, dynamically reconfigure themselves to protect against these threats, and automatically immunize themselves against evolving threat, reducing the response time from days to minutes.", "label": "entailment", "domain": "Networks"}
{"id": "test_1462", "sentence1": "The following figures depict a dataset in which there are multiple peaks of malicious traffic, with increasing rates.", "sentence2": "in this dataset, there are seven periods, with the first period having thirty INVITEs per second, increasing in each peak by an additional ten INVITEs per second.", "label": "entailment", "domain": "Networks"}
{"id": "test_1463", "sentence1": "For example, for anomalous legitimate network traffic, such as unusually high traffic associated with a natural disaster, flash mob, or other unexpected event, the underlying network could automatically grow network resources (e.g.creation of new virtual machines) in order to handle the additional traffic load.", "sentence2": "for anomalous traffic stemming from a malicious attack, the network should dynamically instantiate new firewall rules and/or new virtual security functions in order to block the offending traffic and protect the rest of the net­work from the attack.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1464", "sentence1": "unlike global clustering coefficient, local clustering coefficient measures how a single node is connected with its neighbor nodes (either well connected or loosely connected).", "sentence2": "local clustering coefficient calculation is based on the fraction of the number of present links over the total number of possible links between the node s neighbors as shown in Equation (4).", "label": "entailment", "domain": "Networks"}
{"id": "test_1465", "sentence1": "As the as­sumption of sensor devices equipped on a node, we consider the simplest case in this paper.", "sentence2": "each node is equipped with a sensor device that can observe a single attribute environmental information.", "label": "entailment", "domain": "Networks"}
{"id": "test_1466", "sentence1": "the topology due to the energy exhaustion, the sink has to reconstruct another topology and disseminate the updated information to all nodes.", "sentence2": "the sink estimates how many times the readings can be gathered from all representative nodes using the current topology by simulating communications for the data gathering using energy consumption model which is described in Section 4.1.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1467", "sentence1": "As the number of neighboring nodes increases, such nodes tend to have a large number of children.", "sentence2": "for constructing a topology, the attribute cell energy is considered in the same way as our proposed method.", "label": "neutral", "domain": "Networks"}
{"id": "test_1468", "sentence1": "In [8], the technology of gateway framework was introduced and an experimental result was given of parallel reprogramming in a developed prototype of a central gateway; however, detail of a technical method was not provided.", "sentence2": "in [21], a reprogramming data processing method was provided for time reduction in a central gateway-based network architecture, and the performance of an implemented central gateway was evaluated using a test environment.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1469", "sentence1": "If the DCU owns only one sub-network, then the PCU can transmit the reprogramming data (in the DoIP frame) to the sub-network owned by the DCU, which only identifies the IP.", "sentence2": "if the DCU owns several sub-networks, then multiple socket connections are established dynamically between the PCU and the DCU, and the PCU transmits the reprogramming data for each ECU to the different sub-networks owned by the DCU.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1470", "sentence1": "However, if the DCU owns several sub-networks, then multiple socket connections are established dynamically between the PCU and the DCU, and the PCU transmits the reprogramming data for each ECU to the different sub-networks owned by the DCU.", "sentence2": "when the PCU transmits the reprogramming data of multiple ECUs to sub­networks owned by different DCUs, the PCU transmits the segmented reprogramming data to different DCUs according to each assigned IP number alternatively; these data transfers can then be parallelized in terms of the overall reprogramming.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1471", "sentence1": "Therefore, the total reprogramming time is increased due to the accumulated transmission delay time, and the transmission delay time due to a single buffer usage makes it difficult to accelerate the reprogramming data transfer.", "sentence2": "if a bandwidth of source network (interconnect with the PCU and the DCU) is larger than the bandwidth of a target network (interconnect with the DCU and the target ECU), such as the Ethernet-based source network, the DCU can apply a number of different buffers.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1472", "sentence1": "This means that the total reprogramming time of the parallel and sequential (one-by-one) methods is similar.", "sentence2": "we have restricted the number of ECUs to one per sub-network, and propose multiple ECUs reprogramming on a different sub-network when multiple ECUs are reprogrammed simultaneously in parallel.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1473", "sentence1": "In SPIN, operations were not run in a sequential order but in a randomly chosen way.", "sentence2": "rREQ messages may be received quickly on one side but slowly on another side.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1474", "sentence1": "The proposed method was verified through simulations under all possible circumstances.", "sentence2": "existing routing protocols only support rapid recovery of lost nodes so ad-hoc networks have a problem of significantly low efficiency owing to the lack of support while data is transferred.", "label": "neutral", "domain": "Networks"}
{"id": "test_1475", "sentence1": "As micro systems and sensor technologies have advanced, with recent wireless technologies, autonomous and flexible wireless sensor networks can now be used to send or receive data between moving nodes whereby each node recognizes and measures its own sensor environment and processes data without access points (APs) or base station communication networks, which are generally used in existing infrastructure network wireless technologies.", "sentence2": "ad-hoc network technology enables direct communication between devices so that they can send or receive signals directly between two close users without passing through APs or base station communication networks, thereby effectively providing mobility between nodes compared to existing centralized network technologies.", "label": "entailment", "domain": "Networks"}
{"id": "test_1476", "sentence1": "in Section 2, a routing protocol method used to facilitate communications between nodes in an ad-hoc network is discussed.", "sentence2": "a routing protocol used in an ad-hoc network should respond to frequent changes in states owing to node mobility and lost nodes, as well as being simple to operate and having high reliability.", "label": "entailment", "domain": "Networks"}
{"id": "test_1477", "sentence1": "However, usually the carried information contains confidential data.", "sentence2": "a security aggregation approach is required.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1478", "sentence1": "The state I and state B denote the channel conditions, and p is the transition probability from state I to state B.", "sentence2": "q represents the transition probability from State B to State I.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1479", "sentence1": "This is due to the use of several nonlinear functions, such as the square root, which do not work properly in the Contiki environment due to the lack of support for floating point numbers.", "sentence2": "it is necessary to decrease the number of calculations as much as possible.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1480", "sentence1": "Each application has its own requirement and corresponds to different optimal parameters [16, 15].", "sentence2": "using Neutral Network to learn and narrow the scope of parameters is a practical method.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1481", "sentence1": "Cloud-assisted data broadcasting is an emerging application where cloud computing assists data broadcasting to extend the capacity of system computing and improve the interactiv­ity of the conventional media.", "sentence2": "with the increase in scale, it brings the difficulty on the complexity to provide the sufficient quality of service for diverse receivers.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1482", "sentence1": "Clas­sical proportional scheduling algorithms are vulnerable to ac­cumulated errors because the output is oscillatory and poten­tially unstable when Integral and Derivative components are reduced [12].", "sentence2": "we can conclude that the normal pro­ portional algorithm can meet the criteria of coarse-grained proportions with less consideration on the system stability and the .ner quality of service.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1483", "sentence1": "Figure 16 shows the gap between the observed load and the output from KLEIN. ", "sentence2": "higher levels which need a more global view solve simpler problems at coarser timescales, while the lower levels which need to be more responsive to avoid performance issues can run more rapid reconfigurations.", "label": "neutral", "domain": "Networks"}
{"id": "test_1484", "sentence1": "We thus argue that this INTERMEDIATE design can serve as the basis for a minimally disruptive design for future cellular core architectures that can address today's cellular core limitations.", "sentence2": "nFV is already a reality for carriers [7, 24, 6, 44], and there are many open-source and commercial efforts to virtualize cellular core functions [13, 19, 18, 5, 2, 3].", "label": "entailment", "domain": "Networks"}
{"id": "test_1485", "sentence1": "Our CacheFlow system \"caches\" the most popular rules in the small TCAM, while relying on software to handle the small amount of \"cache miss\" traffic.", "sentence2": "we cannot blindly apply existing cache-replacement algorithms, because of dependencies between rules with overlapping patterns.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1486", "sentence1": "The idea of using DAGs for representing TCAM rule dependencies is discussed in the literature in the context of efficient TCAM rule up­dates [33, 34].", "sentence2": "their aim was to optimize the time taken to install a TCAM rule by minimizing the number of existing entries that need to be reshuffled to make way for a new rule.", "label": "entailment", "domain": "Networks"}
{"id": "test_1487", "sentence1": "Composing two candidate rules to build a cache would simply involve merging their corresponding mixed-sets (and incre-menting appropriate reference counters for each rule) and decomposing would involve checking the reference counters before removing a rule from the TCAM 8.", "sentence2": "figure 10(b) shows that the cache-hit rate for the incremental algorithm is substantially higher as the TCAM size grows towards 2000 rules.", "label": "neutral", "domain": "Networks"}
{"id": "test_1488", "sentence1": "That is, rules R2 and R5 depend on rules R1 and R4, respectively.", "sentence2": "there is a dependency from rule R1 to R2 and from rule R4 to R5.", "label": "entailment", "domain": "Networks"}
{"id": "test_1489", "sentence1": "They do so by building a DAG that captures how different rules are placed in different TCAM banks for reducing the update churn.", "sentence2": "the resulting DAG is not suitable for caching purposes as it is difficult to answer the question we ask: if a rule is to be cached, which other rules should go along with it?", "label": "contrasting", "domain": "Networks"}
{"id": "test_1490", "sentence1": "We can do better than past algorithms by modifying the rules in various semantics-preserving ways, instead of simply packing the existing rules into the available space—this is the key observation that leads to our superior algorithm.", "sentence2": " we “splice” the dependency chain by creating a small number of new rules that cover many low-weight rules and send the affected packets to the software switch.", "label": "entailment", "domain": "Networks"}
{"id": "test_1491", "sentence1": "If the TCAM can store four rules, we cannot select the four rules with highest traffic volume (i.e., R2, R3, R5, and R6), because packets that should match R1 (with pattern 000) would match R2 (with pattern 00*); similarly, some packets (say with header 110) that should match R4 would match R5 (with pattern 1*0).", "sentence2": "rules R2 and R5 depend on rules R1 and R4, respectively.", "label": "entailment", "domain": "Networks"}
{"id": "test_1492", "sentence1": "A key property of the algorithms discussed so far is that each chosen rule along with its mixed (cover or dependent) set can be added/removed from the TCAM independently of the rest of the rules.", "sentence2": "the mixed-sets for any two rules are easily composable and decomposable.", "label": "entailment", "domain": "Networks"}
{"id": "test_1493", "sentence1": "Pre-SDN era declarative networking [17, 16, 19], on the other hand, uses a distributed query engine for fast processing of customized routing protocols, where the database executes routing queries submitted by end-hosts.", "sentence2": "ravel makes the database an active participant that uses views to incorporate multiple high-level and lowlevel abstractions of the network which are orchestrated online.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1494", "sentence1": "In contrast, Ravel makes the database an active par­ticipant that uses views to incorporate multiple high-level and low­level abstractions of the network which are orchestrated online.", "sentence2": "we believe this is likely to be valuable both tech­nically and as a way to encourage more rapid uptake of SDN.", "label": "neutral", "domain": "Networks"}
{"id": "test_1495", "sentence1": "Formally, a network state is consistent if it is compliant with all the application policies.", "sentence2": "consistent network states correspond to applications with invariant-preserving views.", "label": "entailment", "domain": "Networks"}
{"id": "test_1496", "sentence1": "Reliability is achieved by sending each aggregated event packet via multiple paths to the BS.", "sentence2": "drawback of this method is that it is not delay aware.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1497", "sentence1": "From the simulation study it has been observed that mobile base station based data delivery scheme in WSNs may help in reduction of end to end delay and deadline miss ratio.", "sentence2": "we have also observed that packet delivery ratio may decrease due to mobility of base stations.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1498", "sentence1": "This is because it uses a conservative stepwise switch-up and aggressive switch-down strategy to avoid playback interruption, despite the fact that the network bandwidth is sufficient for transmitting a higher resolution segment.", "sentence2": "in the scenario of small bandwidth variation, these results show that RateAdaptation is too conservative to select a higher bitrate segment, and leads to a lower average bitrate and bandwidth utilization.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1499", "sentence1": "Microsoft Media Services (MMS) and Real Time Streaming Protocol (RTSP) are popular solutions used by the video streaming services.", "sentence2": "solutions based on HTTP over TCP are usually preferred owing to the following advantages [8].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1500", "sentence1": "It provides new ways of looking at the structure of a given environment from a real-time perspective based on dynamic up-to-date records of human presence.", "sentence2": "through the day-to-day social activities such as lectures, seminars and regular meetings, we have strong evidence about the existence of .ner-grained relationships as opposed to the user af.liation.", "label": "neutral", "domain": "Networks"}
{"id": "test_1501", "sentence1": "Lee [10] addressed a location problem in three-dimensional space from the application perspective.", "sentence2": "he introduces a 3D coverage location model of Wi-Fi access points (APs) in an indoor environment.", "label": "entailment", "domain": "Networks"}
{"id": "test_1502", "sentence1": "An example of mono-criterion approach is Cost-Benefit Analysis (CBA) [31] which uses, as a single criterion, the Net Present Value [31].", "sentence2": "this approach isn’t always suited for practical problems, as it takes into consideration only one criterion that needs to be expressed in monetary terms. ", "label": "contrasting", "domain": "Networks"}
{"id": "test_1503", "sentence1": "Two roots can be obtained by solving (16) as expressed in Equation (17).", "sentence2": "these techniques are not suitable for inter vehicle distance calculation.", "label": "neutral", "domain": "Networks"}
{"id": "test_1504", "sentence1": "Another GPS-free localization techniques has been presented using Received Signal Strength Indicator (RSSI) [12].", "sentence2": "the localization system has two steps.", "label": "entailment", "domain": "Networks"}
{"id": "test_1505", "sentence1": "Time of Arrival (ToA) based GPS-free localization has been developed [11].", "sentence2": "the GPS-free localization framework operates on two steps.", "label": "entailment", "domain": "Networks"}
{"id": "test_1506", "sentence1": "The term LAMP has retained its underlying rationale also in the 2000s and 2010s Internet, although the contemporary nomenclature allows shuffling; the letter P may currently refer also to Python or Perl, and MySQL may be replaced by PostgreSQL, for instance.", "sentence2": "lAMP is best understood as an idiom for a typical open source software stack for developing and servicing web sites.", "label": "entailment", "domain": "Networks"}
{"id": "test_1507", "sentence1": "If a PHP is detected once for a domain, the remaining preprocessing is skipped for the domain.", "sentence2": "the units of analysis reflect web-servicing domains that supposedly run PHP in at least one corner of their deployments.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1508", "sentence1": "These configuration and modification solutions may prevent a web site from being included in target pools of web scanners and mass-scale exploitation frameworks, but many individual practitioners would likely argue that such solutions rather represent the so-called security-by-obscurity.", "sentence2": "ultimately, (g) applied web intelligence research should be also able to demonstrate a concrete web deployment that successfully utilizes existing big data collections for analyzing real-world questions moti­vated by scholarly hypotheses.", "label": "neutral", "domain": "Networks"}
{"id": "test_1509", "sentence1": "The overall sit­uation is not as grim, however.", "sentence2": "true horror cases are rare; only a negligible amount of identified False web-facing domains ran PHP/3.0 or earlier releases.", "label": "entailment", "domain": "Networks"}
{"id": "test_1510", "sentence1": " A faulty robot follows its assigned trajectory and is indistinguishable from a reliable robot, except that a faulty robot does not detect the target while visiting its location. ", "sentence2": "the target is detected only when its location is visited by a reliable robot.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1511", "sentence1": "Thus, the target is detected only when its location is visited by a reliable robot.", "sentence2": "an algorithm for a parallel search with f faulty robots must make sure that any point on the line can be visited by at least f + 1 robots.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1512", "sentence1": "Thanks to the redundancy, only a subset of the coded chunks is required for reconstructing the original (uncoded) packet.", "sentence2": "packets may be read even when only a part of their chunks is available to read without contention.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1513", "sentence1": "However, it is much easier for hackers to spoof the source IP address as UDP itself is connectionless and does not require a handshake like TCP does.", "sentence2": "it is worthwhile to design a countermeasure against DDoS flooding that suited the DNS traffic.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1514", "sentence1": "Various types of application traverse the mesh network.", "sentence2": "due to mobility and its multi-hop nature, dynamic network configuration is needed to seamlessly support applications for better quality under dynamic network conditions.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1515", "sentence1": "We consider the real 3G connection as our ground truth, which the emulator would ideally mimic perfectly.", "sentence2": "a real 3G connection can significantly vary depending on aspects such as load or scheduling decisions at the network Mean RTT (milliseconds) operator side.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1516", "sentence1": "The reason is that in Wi-Fi and cellular networks the output power levels take values from a discrete set of size N, where N< 100.", "sentence2": "when K = 1, we omit the indices.", "label": "neutral", "domain": "Networks"}
{"id": "test_1517", "sentence1": "For the scope of this paper however we have used average static values, which were taken when the signal strength was full for all 3 interfaces in our lab (indoors) to demonstrate our proposed approach.", "sentence2": "dynamic measurement of bandwidth and signal strength would be critical to further improve the robustness of the proposed solution.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1518", "sentence1": "While the signals were clear on an antenna placed right on top of the phone, we could not see any signal over the body.", "sentence2": "the EM signals generated by the screens of such devices do not propagate well over the body and hence do not cause interference.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1519", "sentence1": "It is based on an intuition that the distance between the two tags is much shorter than the distance from them to the objects in the environment.", "sentence2": "tags at very close positions may share very similar ambient noises and multipath effects.", "label": "entailment", "domain": "Networks"}
{"id": "test_1520", "sentence1": "Theorem 2 dictates that we can always find an optimal frequency combination in a continuous frequency band so that our error tolerance is maximized.", "sentence2": "these systems came with a variety of constraints in order to be extended to a ubiquitous real-time 3D passive device localization system.", "label": "neutral", "domain": "Networks"}
{"id": "test_1521", "sentence1": "6, the front­end antenna reflection loss S11 must be minimized for all operating frequencies.", "sentence2": "it first decides what frequencies are transmitted based on the available bandwidth.", "label": "neutral", "domain": "Networks"}
{"id": "test_1522", "sentence1": "Client devices are unaware of whether content from a domain is available over IPv4 network or over IPv6 network.", "sentence2": "clients must send both AAAA (IPv6) and A (IPv4) DNS queries to their local resolvers to resolve domain names.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1523", "sentence1": "The first are mappings between client and cellular DNS server IP addresses.", "sentence2": "our work precisely describes the current role of a CDN in connecting mobile users to content servers in the evolving cellular ecosystem in the US.", "label": "neutral", "domain": "Networks"}
{"id": "test_1524", "sentence1": "When we run our system at the first time, the running time increases with more unknown clients.", "sentence2": "the running time significantly decreases to 0.065 s after the system's first run, since we only need to estimate the location of the target.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1525", "sentence1": "When a target is running, the CSI changes may vary a lot in a short time.", "sentence2": "liFS can not always get stable CSI changes, causing large errors.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1526", "sentence1": "LiFS is able to localize the two targets when they are located sparsely.", "sentence2": "note that we have CSI from 30 subcarriers and as long as a few of them fall in the clean category, it is enough for our localization purposes.", "label": "neutral", "domain": "Networks"}
{"id": "test_1527", "sentence1": "To remove the noise on raw CSI measurements, we observe that not all subcarriers are affected equally by multipaths even in a rich multipath environment.", "sentence2": "we introduce a novel CSI pre-processing method to filter out those subcarriers greatly affected by multipath and hard­ware noise.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1528", "sentence1": "If many workers are prepared, many context switches and high contention against the two queues occur.", "sentence2": "hoL is not avoidable if there is only one worker.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1529", "sentence1": "Since the only component enqueueing to the input queue is the receiver, flow rate can be controlled to some extent.", "sentence2": "when the receiver enqueues a certain number of requests to the input queue, it requests a context-switch to the scheduler of lightweight threads.", "label": "entailment", "domain": "Networks"}
{"id": "test_1530", "sentence1": "In this scheme, the sendfile system call cannot be used for the file type.", "sentence2": "we adopt the pread system call, which, like the sendfile system call, also preserves offsets in the kernel.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1531", "sentence1": "An alternative approach would be to keep requesting every cache encountered for potentially better related content.", "sentence2": "we believe this might put a high burden on the battery of the UE and the UE-SC traffic.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1532", "sentence1": "Our preprocessing scheme tries to identify the subcarriers not affected by multipath.", "sentence2": "cSIs on the “clean” subcarriers can be modelled and utilized for accurate localization.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1533", "sentence1": "To identify whether a target is located on the LoS path, the key observations are (i) |At| is usually within the range of 4–9 dBm [3] when a human target blocks the LoS path, and (ii) the noise is usually within 1–3 dBm [2].", "sentence2": "a target is more likely located on the LoS path if the averaged CSI change of all subcarriers is larger than 5 dBm.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1534", "sentence1": "To filter out these dirty subcarriers with abnormal CSI changes, our first step stems from the power increase observation at some subcarriers.", "sentence2": "when the CSI amplitude of the k-th subcarrier is increased instead of decreased, we know the subcarrier is affected by multipaths and the CSI measurement at this subcarrier should be filtered out.", "label": "entailment", "domain": "Networks"}
{"id": "test_1535", "sentence1": "We choose to rely on simple local congestion monitoring in Expeditus.", "sentence2": "the only diference is that the first packet of B is a SYN-ACK in this case.", "label": "neutral", "domain": "Networks"}
{"id": "test_1536", "sentence1": "Commodity ToR switches typically have 4.40Gbps uplinks.", "sentence2": "each pod normally has 4 aggregation switches.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1537", "sentence1": "We set the benchmark parameters such that the dominant factor in server performance is the latency introduced by the networking stack at the endpoints and packet processing at VNToR, not by other server or switch bottlenecks.", "sentence2": "we ensure that: the working set of the rules fits in the hardware table; the ASIC-SupE interconnect inside the switch is not saturated;  the servers are neither CPU-nor I/O-bottleneck-ed.", "label": "entailment", "domain": "Networks"}
{"id": "test_1538", "sentence1": "Integrating VNToR in OpenStack merely requires imple­menting a Neutron agent that runs on VNToR s SupE and whose role is to collect, from the Neutron controller, all the relevant security-group state.", "sentence2": "our agent subscribes for changes to: all the local VMs, i.e., those running in the local rack;", "label": "entailment", "domain": "Networks"}
{"id": "test_1539", "sentence1": "We share the general challenge but have a different focus from this work: we design and build a caching system that meets a significantly harder performance baseline, and we achieve this by tailoring the solution to the properties of state-of-the-art datapath memory.", "sentence2": "our agent subscribes for changes to: all the local VMs, i.e., those running in the local rack;  membership information of the security groups where the local VMs belong;  membership information of any security group with which a local VM is allowed to communicate;  all the other relevant network-wide translations defined by Neutron, e.g., encapsulating tunnels.", "label": "neutral", "domain": "Networks"}
{"id": "test_1540", "sentence1": "The ASIC handles all traffic that can be served from the hardware table and passes the rest to the software for­warders.", "sentence2": "the moment the rule is promoted to the hard­ware table, there is packet reordering (directly proportional to the amount of buffering in the software forwarder and the hardware-table update latency tup), but throughput climbs to line rate within 20ms.", "label": "neutral", "domain": "Networks"}
{"id": "test_1541", "sentence1": "In order to evaluate the T-SIMn framework, we use an evaluation methodology similar to that used to evaluate T-RATE [2].", "sentence2": "we conduct an experiment (which produces a trace) using a round-robin ordering of rate configurations and then in SIMn we conduct a simulation using a round-robin ordering that differs from the experiment.", "label": "entailment", "domain": "Networks"}
{"id": "test_1542", "sentence1": "As real devices are used, the MAC and parts of the physical layer are not simulated.", "sentence2": "sig­nal propagation is simulated using the FPGA to alter the signals being transmitted between devices.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1543", "sentence1": "In 802.11n networks, the fate of one frame can impact the number of frames that can be aggregated in the next frame due to the Block-Ack Window (BAW).", "sentence2": "upon closer investigation, we realized that the simulator is in fact accurate and that the problem was with the methodology used to evaluate the accuracy of the framework.", "label": "neutral", "domain": "Networks"}
{"id": "test_1544", "sentence1": "During trace collection, rates are grouped by a combina­tion of the Guard Interval (GI) and the CB.", "sentence2": "this scenario is comprised of a mix of carrying the iPhone at walking speed and standing still in an office and hallway environment as explained in Section 4.", "label": "neutral", "domain": "Networks"}
{"id": "test_1545", "sentence1": "This suggests that SIMn accurately handles rate configurations using different physical layer transmis­sion features.", "sentence2": "sIMn accurately calculates the transmission time of a frame (including ACK and DCF timing) for the combinations of the physical layer features shown.", "label": "entailment", "domain": "Networks"}
{"id": "test_1546", "sentence1": "To increase the realism of emulators, hybrid approaches using traces to simulate the physical layer and emulation for the MAC layer have also been proposed [13, 24], however these have been limited to 802.11b networks and rely on measurements of Signal-to-Noise Ratio (SNR) to simulate the physical layer.", "sentence2": "the SNR can not be used to accurately predict frame fates [9].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1547", "sentence1": "Being able to accurately simulate A-MPDUs of any length using frames collected using only A-MPDUs of maximum length is the  key insight and critical requirement to enable trace-based simulation of 802.11n networks. ", "sentence2": "we now evaluate SIMn’s accuracy when simulating the throughput of frames aggregated with fewer subframes during simulation than were obtained during trace collection.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1548", "sentence1": "We showed that such peaky signals also have, in the wide-band limit, very strong AJ properties [9, 10].", "sentence2": "we showed that, through the use of peaky signals, energy-limited jammers do not affect capacity in the wideband regime.", "label": "entailment", "domain": "Networks"}
{"id": "test_1549", "sentence1": " Despite the impact of the traffic arrival distributions on the goodput performance, there is no impact on the goodput-efficient routing strategy.", "sentence2": "multi-hop routing under moderate traffic loads and direct transmission under light and heavy traffic loads is the optimum choice for enhancing goodput performance in multi-hop wireless networks, independent from the traffic arrival distribution, generalizing the results in [5, 23] for the considered traffic arrival distributions.", "label": "entailment", "domain": "Networks"}
{"id": "test_1550", "sentence1": "We are focused on the feasibility of virtualization implemented as a user-level program so that the result may work on top of a diverse mix of underlying system stacks.", "sentence2": "we are intrigued by the possibility that a specially designed P4 program itself may provide the benefits of programmable data plane virtualization by emulating other P4 programs.", "label": "entailment", "domain": "Networks"}
{"id": "test_1551", "sentence1": "HyPer4, however, moves most of the parsing decision logic into the ingress pipeline, and the actual parse graph for HyPer4 simply extracts a specified number of bytes without any higher level structure.", "sentence2": "hyPer4 makes an end run around a restriction normally imposed by P4, for better or for worse.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1552", "sentence1": "HyPer4 can preemptively declare statically bound registers, counters, and meters for every table that emulates the match piece of a match-action stage, but this approach is likely to prove infeasible for many hardware P4 targets.", "sentence2": "registers can vary in width, which for the preemptive ap­proach would require HyPer4 registers be declared with sufficient width to cover the maximum need, resulting in a lot of wasted register memory for the average case.", "label": "entailment", "domain": "Networks"}
{"id": "test_1553", "sentence1": "Figure 1 illustrates a packet exchange procedure of the proposed scheme as an example where source (S), destination (D), and two cooperators (C1, C2) exist.", "sentence2": "by doing this, we can verify the advantages of the cooperative method in terms of the system throughput.", "label": "neutral", "domain": "Networks"}
{"id": "test_1554", "sentence1": "The lower data transmission power is used, in other word, the larger IR of R is, the worse the LIRC problem becomes.", "sentence2": "as a solution for the LIRC problem, we try to narrow down the interference range within RTS/CTS transmission range by allocating more power.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1555", "sentence1": "Since the interferer doesn’t receive RTS or CTS and thus cannot enter the sleep state, it may cause collisions with data at R. ", "sentence2": "in this paper, we propose a multi-channel based MAC protocol with power control to solve LIRC problem by separating the channels for control and data packets and allocating appropriate power for data transmission in underwater environment.", "label": "neutral", "domain": "Networks"}
{"id": "test_1556", "sentence1": "In this case, all links exhibit fairly good performance for both SL1 and SL2.", "sentence2": "with the lower source level SL1, the phono­absorbing surfaces tend to suppress the majority of the sec­ondary paths: this translates into high PDR for almost all links, and in an improvement of 2 .", "label": "entailment", "domain": "Networks"}
{"id": "test_1557", "sentence1": "A specific case of technique i) above is the addition of metal dust to butyl rubber [17]: this achieves a better matching with the water impedance and increases the attenuation of sound energy in the material.", "sentence2": "this type of rubber tends to absorb water over time, which degrades its phono­absorbing properties.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1558", "sentence1": "These environments provide characteristics more akin to shallow water scenarios, giving the research team more insight as to what to expect in a real environment, and thus making it more ready for the sea experiment.", "sentence2": "large testing pools, such as the one employed by CMRE [7], require a large infrastructure which may not be available to smaller research laboratories, e.g., as typically found in universities.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1559", "sentence1": "Secondary paths have a much more limited effect after applying the coating.", "sentence2": "this is not the only impact of the phono-absorbing coating.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1560", "sentence1": "The SNs are a battery powered devices, and they consume their energy during the network process.", "sentence2": "the remaining energy can be used as a metrics for selecting the CHs.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1561", "sentence1": "Like, if we have setup CSr in one of the bedrooms, then it may be possible that mediators of drawing room cannot pass data to CSr directly.", "sentence2": "a data aggregation algorithm is required to aggregate all data in the network at CSr.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1562", "sentence1": "Data encryption and anonymization are the most common methods to protect private information for the perceptual layer of IoTs.", "sentence2": "the connection of physical things to the Internet makes it possible to access remote sensor data and control the physical world from a distance [2].", "label": "neutral", "domain": "Networks"}
{"id": "test_1563", "sentence1": "Based on the technology of the Internet, there are many advanced methods to keep an IoTs system connect safely with external network.", "sentence2": "these traditional technologies are not suitable for the perceptual layer of IoTs.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1564", "sentence1": "With Interest filtering, we could enable lights to provide service on multiple meaningful names with a single FIB entry.", "sentence2": "the current Interest filtering is only supported by NDN client applications.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1565", "sentence1": "To ensure the execution of the command, each light node is required to reply with an ACK Data by appending its own light ID to the received Interest name.", "sentence2": "the current Interest filtering is only supported by NDN client applications. ", "label": "neutral", "domain": "Networks"}
{"id": "test_1566", "sentence1": "Software­based discrete event simulators are based on a simula­tion clock rather then real time.", "sentence2": "the required computation time increases dramatically with the num­ber of simulation nodes.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1567", "sentence1": "If the host is ILNPv6capable, new DNS records for NID and L64 values will be returned, or the lookup in /etc/hosts will return an I-LV from a /etc/hosts file that has extended syntax as shown in Figure 3.", "sentence2": "iLNP can offer multihomed IP connectivity as first class functionality so does not suffer the same issues, e.g. security issues, that are currently of concern for MP-TCP [31].", "label": "neutral", "domain": "Networks"}
{"id": "test_1568", "sentence1": "Expanding variety of end devices connected to the Internet has introduced high demand to flexibly authenticate and grant them the necessary access to the network.", "sentence2": "it is not realistic to expect of all the end devices, including less capable and low-cost devices like sensors or embedded systems, to satisfy the requirement of integrated authentication procedure like 802.1x.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1569", "sentence1": "Note that the nodes in S are not candidates for the MCC as they represent the attackers (they are not vulnerability options for the attackers to exploit).", "sentence2": "the nodes in T are eligible as they represent exploitation opportunities for the attackers on critical targets.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1570", "sentence1": "These graphs were instantiated from an attack template representing an ICS secured to industry standards.", "sentence2": "each link in our instantiated attack graphs represented actual attack types that occur between the modeled services/entities.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1571", "sentence1": "The exact solution clearly grows exponentially.", "sentence2": "each node represents a distinct vulnerability at a particular location (usually on some network host).", "label": "neutral", "domain": "Networks"}
{"id": "test_1572", "sentence1": "To calculate width, we must find the smallest set of colors such that at least one color in the set will occur on every possible attack path.", "sentence2": "layer i contains all the nodes that are distance i from the set of attacking nodes.", "label": "neutral", "domain": "Networks"}
{"id": "test_1573", "sentence1": "Once this is done, it is free to run Docker-packaged applications in the VM environment.", "sentence2": "there is no performance bene.t for Inter-VM Inter-Container-1Layer, which has the similar performance to the default mechanism (denoted as *-Def).", "label": "neutral", "domain": "Networks"}
{"id": "test_1574", "sentence1": "The focus of this simula­tion framework is to optimize WSN deployments regarding power efficiency and reliability.", "sentence2": "a model of each WSN node, including its function, timing, and power consumption needs to specified, where the detail granularity of the model can be adapted depending on the simulation aspect that the user is interested in.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1575", "sentence1": "The real-world wireless bridge scenario, where the WARP nodes use their antennas for Single Input and Single Output (SISO) communication.", "sentence2": "the measurement setup is shown in Fig. 9(a).", "label": "neutral", "domain": "Networks"}
{"id": "test_1576", "sentence1": "Besides the memory usage, an important factor for the scalability of our approach is the required computation time, as, in DES, the simulation time does not necessarily match the actual time to execute the simulation, which is influenced by many factors.", "sentence2": "we measure the computation time for a simulation of 1 s in networks of different sizes.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1577", "sentence1": "The execution of WARP instances in different processes, which are coordinated via Inter-Process Communication (IPC), al­lows the simulation of small networks.", "sentence2": "this approach does not scale well for larger typologies as each additional WARP node requires its own process.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1578", "sentence1": "CoWS only allocates the amount of memory that a simulated WARP node currently needs.", "sentence2": "we are able to reduce the memory allocation per WARP node to about 4 MB.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1579", "sentence1": "For embedded devices, in contrast to systems that are managed by an OS, it is typically possible to write to any addressable part of the memory. ", "sentence2": "writing to arbitrary parts of the memory may lead to unexpected and erroneous behavior of the device, which is, in addition, difficult to track down.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1580", "sentence1": "Incorrect signatures make DNSSEC resolvers fail during the verification process. ", "sentence2": "an attacker on the path upstream from a validating resolver can cause signature validations to fail by modifying traversing DNS packets.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1581", "sentence1": "Meanwhile, ISPs formed a hierarchical structure and were classified by tiers, with higher tier ISPs cover larger geographic regions and provide transit service for lower tier ISPs.", "sentence2": "large content providers, e.g., Google, are deploying their own wide-area networks so as to bring content closer to users and bypassing Tier-1 ISPs on many paths, which is known as the flattening phenomenon of the Internet topology", "label": "contrasting", "domain": "Networks"}
{"id": "test_1582", "sentence1": "Looking at both situations it is obvious that the pass to Costa is more likely to lead to a shooting chance (i.e., higher reward) but is inherently more risky and requires more skill to execute than the pass to Fabrega.", "sentence2": "current passing measures assign both passes with the same weighting (1 for successfully making the pass and 0 for not)", "label": "contrasting", "domain": "Networks"}
{"id": "test_1583", "sentence1": "The extracted fields are then used to perform Access Control List (ACL) or forwarding table lookups in later stages to support packet classification or routing functions, e.g., a firewall makes a decision on whether to drop a packet according to a standard TCP/IP five-tuple extracted by the parser.", "sentence2": "traditional packet parsers are designed for the specific network environment, which are too rigid to accommodate new protocols [1].", "label": "contrasting", "domain": "Networks"}
{"id": "test_1584", "sentence1": "Clearly, multi-radio gateways can serve as translators among heterogeneous wireless devices with incom­patible physical layers.", "sentence2": "this solution is inherently limited in several aspects.", "label": "contrasting", "domain": "Networks"}
{"id": "test_1585", "sentence1": "For example, in ZigBee devices, the packet-level RSSI information is sampled at only 31.25KHz, while the phase­shift is obtained at 4MHz by the ZigBee PHY for (de)modulation.", "sentence2": "the throughput of packet-level CTCs is inherently bounded by a low sampling rate (e.g., KHz) in contrast to hundreds of Kbps for native ZigBee communication.", "label": "entailment", "domain": "Networks"}
{"id": "test_1586", "sentence1": "Without hardware modification, the WiFi signals transmitted in the pilot/null subcarrier cannot be controlled by software.", "sentence2": "if the pilot subcarriers overlap with the frequency bands of ZigBee devices, WEBee cannot work properly.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1587", "sentence1": "These constellation points are controlled by the source bits of the WiFi payload.", "sentence2": "source bits in a selected WiFi payload determine the QAM constellation points after the WiFi modulator.", "label": "entailment", "domain": "Networks"}
{"id": "test_1588", "sentence1": "Compared with the 16µ duration of a ZigBee symbol, a WiFi symbol occupies 4µs.", "sentence2": "a full ZigBee symbol has to be segmented before emulated by four individual WiFi symbols , as shown in Figure 10.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1589", "sentence1": "The key idea of MaxSNR-MinHop is that the routing tables are constructed for a set of discrete SNR thresholds.", "sentence2": "we relax our target to maximize the signal strength received at intended node.", "label": "neutral", "domain": "Networks"}
{"id": "test_1590", "sentence1": "If the G/L bit is 1, it is known that the BSSID is used as a local address in closed network.", "sentence2": "this research does not target such BSSID showing the local address.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1591", "sentence1": "The ratio of existing decreases tem­porarily but it starts to increase again as the total observed days increases.", "sentence2": "it is concluded that eight (the ob­served days) is the change-point since hotspots are elimi­nated and the amount of data is not too limited.", "label": "reasoning", "domain": "Networks"}
{"id": "test_1592", "sentence1": "According to Wi-Fi radio wave characteristics, RSSI showing more than -75[dBm] are in high proportion to its distance from the RBS while RSSI less than showing -75[dBm] is in low proportion.", "sentence2": "this research targets the RBS showing RSSI more than -75[dBm].", "label": "reasoning", "domain": "Networks"}
{"id": "test_1593", "sentence1": "We run the program on the PC whose OS is Windows 8.1 Pro, proces­sors are Intel (R) Core (TM) i7-4770CPU @ 3.40GHz, and RAM is 24.0GB.", "sentence2": "this research targets the RBS showing RSSI more than -75[dBm].", "label": "neutral", "domain": "Networks"}
{"id": "test_1594", "sentence1": "However, variance converges to around nine regardless of the times or days.", "sentence2": "in addition, in the process of modeling GMM[2], they can be eliminated so this research does not target to detect mobile hotspots.", "label": "neutral", "domain": "Networks"}
{"id": "test_1595", "sentence1": "We believe that a similar integration will be necessary in order for VLC to take off and, accordingly, the packaging and integration of VLC hardware within luminaires and wireless devices is a key challenge that needs to be addressed.", "sentence2": "such integration and product readiness will force design decisions that are less expensive, less complex, and more robust – likely at the expense of overall communications performance.", "label": "entailment", "domain": "Networks"}
{"id": "test_1596", "sentence1": "In our first experiment, we analyze whether the intuition sketched in Figure 1 holds.", "sentence2": "we study whether switching to a receive beampattern that is not the one that provides the highest gain towards a transmitter helps mitigating lateral interference.\n", "label": "entailment", "domain": "Networks"}
{"id": "test_1597", "sentence1": "Faulty nodes or parallel operation of incompatible standards which distort the channel, may cause signi\u001bcant interference not just via main lobes but also via side lobes. ", "sentence2": "the best beampattern for a receiver may not be the one whose main lobe most accurately points towards the transmitter but the one whose side lobes do not capture any interference.", "label": "entailment", "domain": "Networks"}
{"id": "test_1598", "sentence1": "To overcome interference or signal jamming, we aim to search for an alternative beampattern that (i)maximizes the antenna gain towards the intended direction and (ii)minimizes the impact of interference.", "sentence2": "an alternative beam should have a high correlation with the currently used beampattern but different zeros to steer the antenna away from the interference direction.", "label": "entailment", "domain": "Networks"}
{"id": "test_1599", "sentence1": "The baseline audio features provided by DSC consist of 5 formant features and 74 covarep features. ", "sentence2": "formant features consists of first 5 formants, covarep features consists of prosodic features, voice quality features and spectral features. ", "label": "entailment", "domain": "Networks"}
{"id": "test_1600", "sentence1": "At the same time we transform abstract states, we perform verifications to prove that every C#minor expression evaluates safely (without blocking) in its evaluation context.", "sentence2": "we check that every load and store is performed within bounds and with the correct alignment.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1601", "sentence1": "Therefore, it seems necessary to compare two channels, or at least one channel and one abstract state.", "sentence2": "channels being records of functions, comparison is not decidable.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1602", "sentence1": "Our solution is to maintain the invariant that channels never contain more information than what is con­tained in the abstract values they are paired with.", "sentence2": "at the top of the combination of domains, when we manipulate a pair (ab, chan) of an abstract value and a channel, we will make sure that γ(ab) ⊆ γ(chan) holds.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1603", "sentence1": "Analyzing mathematical, exact integers instead of machine integers greatly simplifies the implementation and proof of abstract integer operations.", "sentence2": "there is no benefit in analyzing FP numbers using a more mathematical type such as rationals: FP operations (with rounding) behave well with respect to FP ordering, and abstract operations that use rationals are costly.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1604", "sentence1": "Section 2 presents the gen­eral architecture of the analyzer.", "sentence2": "the returned channel contains no more information than what is contained in the returned abstract value and the given channel.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1605", "sentence1": "On the example of intervals above, our construction is very close to wrapped intervals [31].", "sentence2": "computing (optimal) least upper bounds between sets of symbolic equalities is known to be dif.cult [21], so we settle for an overapproximation.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1606", "sentence1": "We use d to denote an implementation of a GC algorithm when the context is clear.", "sentence2": "if d = 1, then GRA always selects the sealed block containing the smallest number of valid pages for GC, and we call it the GREEDY algorithm.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1607", "sentence1": "As d increases, both the impact of clustering and skewness on the cleaning cost vanishes.", "sentence2": "data locality within a workload, including both clustering and skewness, has a signi.cant impact on GREEDY, but has no impact on RANDOM.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1608", "sentence1": "Even though skewness worsens the cleaning performance of GREEDY for locality-oblivious GC (see Figure 6(b)), it improves the performance of locality-aware GC.", "sentence2": "incor­porating locality awareness into GC design is more signifi­cant for workload with higher skewness.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1609", "sentence1": "Data locality affects the performance of locality-oblivious GC with a given implementation (i.e., GRA with a fixed d): (1) cleaning cost increases when either the active region size or the skewness increases, (2) the increase is more pronounced for a smaller window size.", "sentence2": "data locality has the most significant impact on GREEDY, but has no impact on RANDOM.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1610", "sentence1": "Note that the design strategy of data grouping exploits data locality by eliminating the skewness within a work­load.", "sentence2": "it separates a highly-skewed workload into multiple uniform workloads with no skewness so as to re­duce the cleaning cost (see §5.2 for the impact of skewness).", "label": "entailment", "domain": "Hardware"}
{"id": "test_1611", "sentence1": "Since it is a general consensus that cleaning cost of GC algorithms may be reduced when data can be differenti­ated [11, 14] and a workload includes n + 1 types of data pages, it is of interest to study the performance of GC that exploits data locality using data grouping, which differenti­ates different types of data pages and stores them separately in different regions in an SSD.", "sentence2": "we analyze how data grouping influences the performance of GC in SSD, and how much is the influence for workloads with different degrees of locality?", "label": "entailment", "domain": "Hardware"}
{"id": "test_1612", "sentence1": "All page writes, such as external writes (due to workload) and internal writes (due to GC), are directed to the write frontier, and pages are sequentially written to the write frontier.", "sentence2": "we formulate our workload model based on our trace anal­ysis as follows.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1613", "sentence1": "In addition, DTR introduces new pipelined cells which may allow optimizations using retiming.", "sentence2": "according to the fault-model SET (1 , K ), no error occurs within K clock cycles after the last error.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1614", "sentence1": "The maximum enumeration number in all our experiments will not exceed 104 and enumeration-based approach will .nish within a second.", "sentence2": "we just use the enumeration-based approach in our experiments.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1615", "sentence1": "Under such circumstances, loops in an antichain may execute in parallel.", "sentence2": "loops in an antichain will not share compo­nents with conservative scheduling.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1616", "sentence1": "Section 6 presents the efficiency of the proposed throughput optimization algorithm over traditional approaches with experimental results and the conclusions are presented in Section 7.", "sentence2": "loop pipelining [17] is a key optimization technique in highlevel synthesis.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1617", "sentence1": "Multiplexers for resource sharing are non-shareable component candidates.", "sentence2": "since the area of shareable component candidates are signicantly larger than the multiplexers by definition, in this paper, we made a simplification by assuming that the area of non-shareable component candidates is irrelevant to the resource allocation vectors.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1618", "sentence1": "A well known hardware solution to perform stride permutation in bitonic sorting is the delay feedback or delay commutator module widely used in FFT designs [10, 11].", "sentence2": "using the delay feedback or delay commutator for sorting needs the inputs to be fed in with some particular temporal order.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1619", "sentence1": "These works either improve the sorting algorithms in terms of throughput and latency, or adapt the algorithms to a variety of general purpose parallel architectures such as SIMD or MIMD machines.", "sentence2": "when considering both energy and performance as the key metrics, hardware-based sorting solutions are preferred.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1620", "sentence1": "In the second stage of SPN, each memory block can be implemented with single-port memory to permute a single data sequence.", "sentence2": "when processing continuous data streams, dual-port memory is required as concurrent read and write access to different memory locations need to be performed.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1621", "sentence1": "Similarly, applications deployed on ASICs and FPGAs are often hindered by slow external memories.", "sentence2": "to achieve good performance, hard­ware designers must optimize main memory usage.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1622", "sentence1": "For example, the memory sub-system for the Laplace benchmark is likely very similar to what one would select manually.", "sentence2": "some of the memory subsystems are logical, but would likely re­quire manual experimentation to discover.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1623", "sentence1": "This will be a short path, as the exception handling conditions are largely calculated before the final output.", "sentence2": "abstracting with credit is permitted.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1624", "sentence1": "Like an operating system, LEAP provides a unified layer of abstraction on top of device-specific drivers that interface the underlying FPGA device, on-board memory and the host system an FPGA card is plugged into.", "sentence2": "our setup uses LEAP's scratchpads (SPs), a memory interface abstraction for FPGA applications.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1625", "sentence1": "The concept of shadow registers is shown in Figure 1.", "sentence2": "both the input and output of the user register must be accessible; the input is fed into a shadow register, and the outputs of both the user and the shadow register are compared using an XOR gate.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1626", "sentence1": "Whilst, on the surface, it would appear that constraining the insertion process to use only spare resources that were left behind may be overly restrictive, significant flexibility is recouped from exploiting the convenient property that debugging signals can be connected to any trace-buffer input for it to be observable.", "sentence2": "unlike the user circuit, where nets need to be routed exactly from a single source pin to a predetermined set of sink pins, debug nets need only be routed from any point along the existing user net to any one of the many trace-buffer sinks that are available.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1627", "sentence1": "Another area that we would like to pursue is to explore ways to effectively insert the violation detector (XOR) logic which compares the value in the user register with the shadow register.", "sentence2": "this detection logic (and any downstream infrastructure) has the freedom to be placed anywhere on the device, as long as it does not extend Tcrit ; however, because this operation is a pairwise reduction (i.e.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1628", "sentence1": "If we did try to do this, placement would fail due to particular characteristics of the device chosen rather than anything inherent in the CLB architecture.", "sentence2": "the area constraint is only applied on the CLB resources.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1629", "sentence1": "Instead, we interpret their waveforms as two probability density functions (PDF) or histograms of their sample index.", "sentence2": "fortunately, Cypress provides an associ­ated Microsoft Excel spreadsheet for power consumption in the families of various PSoC devices.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1630", "sentence1": "In digital domain, it is quite easy to perform such a task by using logic comparators and integer counters, which together behave like a histogram builder logically.", "sentence2": "today, even with some of the most remarkable improvements in digital computing, many digital signal processing algorithms still can not achieve real-time performance within the power budget constraints required for many portable applications.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1631", "sentence1": "As shown in Figure 15, for a conventional FPGA-based convolver, the execution time increases with input vector size almost exponentially, which indicates that the execution time will be exceedingly long when the input vector is long.", "sentence2": "once again, all circuit components in Figure 8, except all necessary resistors, can be readily found in our targeted programmable analog device, PSoC.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1632", "sentence1": "It is well known that analog signals are far more sensitive to noise than digital signals because digital quantization, binary numbering, and Boolean algebra jointly increases error margins during computing.", "sentence2": "analog signals are represented as continuous voltages or currents, whose precision can be affected directly if any noises occur.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1633", "sentence1": "Data processing is completed much faster compared to that done by a low-power micro-controller unit that processes data in a sequential manner [1].", "sentence2": "the configuration was done by setting the designated 140 atom switches (or 70 CASs) to the ON state in\neach cell.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1634", "sentence1": "time programming memory allows avoidance of the external ROM and power efficient wakeup.", "sentence2": "this has the drawback of a high programming voltage of more than 5 V, resulting in relatively lower scaling capability and higher operating power.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1635", "sentence1": "MyHDL is already a powerful language for hardware devel­opment.", "sentence2": "we expect the per-device fault rates in an exascale computer will be similar to those observed in todays DRAM devices.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1636", "sentence1": "On the other hand, MyHDL has great support for parameterization.", "sentence2": "myHDL did not have support for converting code that used attributes, so abstraction was limited.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1637", "sentence1": "However, they are simply outdated; poor parameterization limits high level design and modern abstraction features such as classes are missing.", "sentence2": "myHDL has great support for pa­rameterization.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1638", "sentence1": "The worst-case responses times calculated by the analysis tool SymTA/S are specific to the timing analysis model obtained after transformation of the MyCCM design model.", "sentence2": "some calculated response times may be related to tasks and operations resulting from the splitting process explained in Section 3.2.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1639", "sentence1": "Operation calls in MyCCM are namely either synchronous (blocking) or asynchronous (non-blocking).", "sentence2": "a major reason is the lack of engineering methods allowing the integration of the formal timing analysis in the different phases of the development process of real-time embedded systems.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1640", "sentence1": "Prior work has shown that DRAM vendors maintain an approximately constant fault rate per device across technology generations, despite reduced feature sizes and increased densities [9].", "sentence2": "we expect the per-device fault rates in an exascale computer will be similar to those observed in todays DRAM devices.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1641", "sentence1": "The wires on this bus are shared from the memory controller to each DIMM s register.", "sentence2": "errors on these wires will be seen by all DRAM devices on a DIMM.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1642", "sentence1": "The figure makes clear that SRAM fault rates are dominated by faults in the L2 and L3 data caches, the two largest structures in the processor.", "sentence2": "it is clear that faults occur in many structures, including smaller structures such as tag and TLB arrays.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1643", "sentence1": "The focus of this work is to provide insight and guidance to system designers and operators on characteristics of a reliable system.", "sentence2": "our focus is on analyzing the efficacy of existing hardware resilience techniques, their impact on reliable system design, and whether they will be adequate for future systems.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1644", "sentence1": "We note in passing that rIOMMU relies on the predictability of the order of IOVA (un)mappings not the order by which the IOVAs are used by the device.", "sentence2": "so long as IOVAs are valid (mapped), they can be used out of order.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1645", "sentence1": "Therefore, the existing IOMMU allows the I/O device to access an already unmapped buffer if the page in which it resides additionally houses a still-mapped buffer.", "sentence2": "rIOMMU eliminates this type of vulnerability.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1646", "sentence1": "This property is appealing because different target buffers can, and often do, reside on the same page.", "sentence2": "the existing IOMMU allows the I/O device to access an already unmapped buffer if the page in which it resides additionally houses a still-mapped buffer.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1647", "sentence1": "Importantly, ring semantics dictate that (1) the driver work through the ring in order, one descriptor after the other, and that (2) the I/O device process these descriptors in the same order.", "sentence2": "iOVAs are short-lived and the sequence in which they are used is linearly predictable: each IOVA is allocated, placed in the ring, used in turn, and deallocated.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1648", "sentence1": "Write atomicity: For write atomicity, DeNovoSync0 sim­ply uses a single-reader protocol for synchronization (at the word granularity, which is the coherence granularity for De-Novo).", "sentence2": "a synchronization read is always required to register itself at the LLC and only one read can be registered at a time.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1649", "sentence1": "Correspondingly, the MLUs need to read f X 16 bits of centroids and u X f X 16 bits of test­ing instances at each cycle.", "sentence2": "the read width of centroids and testing instances are different.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1650", "sentence1": "In different rounds of counting, the same training instance could also be reused, but the reuse distance depends on data characteristics (instead of algorithm characteristics), thus is not deterministic. ", "sentence2": "we cannot expect a predefined tiling strategy that can reduce the corresponding memory accesses.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1651", "sentence1": "Kernel matrix computation shares a similar locality property to distance calculations in k-NN, except that for each pair of instances, kernel matrix computation computes the value of kernel function instead of computing the distance.", "sentence2": "we reuse the tiling mechanism designed for distance calculations.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1652", "sentence1": "The maximum Nmt of the sequential scheme case is fixed at gt +1 , regardless of the number of attempts of its proof generation process.", "sentence2": "as the number of attempts of the process increases, the probability of the process completion also increases, and the time for the proof generation process with error control is faster than the proof generation process without error control.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1653", "sentence1": "In Juels protocols of [6], a pair of tags generates its replies until the timeout in order to limit the time for the yoking proof generation process.", "sentence2": "the timeout occurs in each tag, independently.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1654", "sentence1": "One can observe a significant reduction of negative slacks at the primary outputs after optimization.", "sentence2": "the number of primary outputs with negative slack reduced from 2004 to only 2 (i.e., more than 99% of reduction).", "label": "entailment", "domain": "Hardware"}
{"id": "test_1655", "sentence1": "The priority of a location corresponds to its distance to the location as­signed by Algorithm 1.", "sentence2": "the more expressive reductions are obtained on circuit leon3: 11.89% on wns and 23.20% on tns.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1656", "sentence1": "on average, our technique further improves wns and tns by 13.42% and 20.18%, respectively.", "sentence2": "the more expressive reductions are obtained on circuit vga lcd: 50.22% on wns and 89.52% on tns.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1657", "sentence1": "Taken the input and output TVFDs into account, each FF can find a feasible region for movement without violating timing, which we label as timing-violation-free region (TVFR) in Figure 2.", "sentence2": "clustering a few single-bit FFs with overlapped TVFRs can optimize power and area without violating timing constraint.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1658", "sentence1": "Then the group of AB and the group of CD will be clustered into a 4-bit group.", "sentence2": "the basic idea is to propose an exact formulation of the number of clusters based on the Dirac delta function, and smooth it using the Gaussian function to make it applicable in a nonlinear programming (NLP) framework with another objective in wirelength.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1659", "sentence1": "The solution after one step of the Nesterov s method, at line 4, may violate the timing constraint.", "sentence2": "we apply a projection step in line 5 to find a closest solution in the feasible space to replace the intermediate infeasible solution.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1660", "sentence1": "As is shown in Figure 5(a), when FFi is adjacent to less than 3 FFs, FFi will attract more FFs to form a cluster of size 4; when there are more than 3 neighbors, FFi will repel the extra FFs.", "sentence2": "it guarantees the maximization of 4-bit clusters number.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1661", "sentence1": "For example, two resilient processors can work at the same frequency, however due to process variation, one may have more timing errors than the other when running the same benchmark program.", "sentence2": "the metrics of resilient system binning are more complicated than traditional binning.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1662", "sentence1": "Fig. 1(d) finds nearest legalization spot for each cell.", "sentence2": "it generates less cell displacement.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1663", "sentence1": "Only then (in a third step) will we be able to tackle the task of analog layout synthesis (i.e., fulfilling requirements) in a com­prehensive and consistent manner.", "sentence2": "analyzing and verifying capabilities are a precondition for synthesizing [4] (Fig. 2).", "label": "entailment", "domain": "Hardware"}
{"id": "test_1664", "sentence1": "The purpose of this paper is to give an up-to-date overview of ana­log design automation, highlighting physical design, its specific characteristics and its current research areas from both an industrial and an academic perspective.", "sentence2": "we will first review the analog layout design problem itself and discuss various aspects of today's design flows.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1665", "sentence1": "Some promising approaches are described below.", "sentence2": "procedures re-use expert knowledge with the result of solutions previously conceived and captured in a procedural de­scription by a human expert, thus imitating the expert's decisions in a straight-forward manner.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1666", "sentence1": "Some reasons are that (1) the design is too application-specific, (2) even small changes to the circuit may require large changes in layout, (3) a new technology node is used, and (4) the shape of a layout module does not fit.", "sentence2": "careful consideration reveals that the underlying reason is that the layout view does not encompass any remaining degrees of free­dom.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1667", "sentence1": "This shortfall is primarily rooted in the analog IC design problem itself, which is very much more complicated even for small problem sizes: it deals with a large number of specific circuit classes; it requires a customized design approach for each circuit class; and analog circuits are very susceptible to noise and process variations.", "sentence2": "the work and costs involved in producing analog layout is a serious bottleneck in IC design, despite numerous attempts at automating the process.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1668", "sentence1": "If an I/O device supports single-root I/O virtualization (SRIOV), a VM is able to issue I/O instructions directly to the device in a way that is isolated from other VMs, and therefore VM exits are avoided when I/O instructions are issued.", "sentence2": "this is achieved through careful installation of timer interrupts that the hypervisor sets up on a dedicated core and migration of timer interrupts that a VM sets up when the VM is suspended and migrated.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1669", "sentence1": "Their signal is also strongly affected by motion noise and usually requires complex filtering/preprocessing steps.", "sentence2": "we achieve an error rate as low as 7% (based on eye motions alone) for the word count estimation (std = 0.5%).", "label": "neutral", "domain": "Hardware"}
{"id": "test_1670", "sentence1": "This observation about pure movement and fine motor parts of the overall task has an impact on the Fitts’ index of performance, and index values (3) and (4) may signi.cantly dif­ fer, depending on the task.", "sentence2": "the observation that the overall task can be divided into different parts implies that those indices are not competing - they are simply conveying different information.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1671", "sentence1": "In our implementation of the telerobotic Fitts' task, we consider two types of pegs, \"thick\" and \"thin\", with widths respectively equal to 8.00 and 4.60 millimeters.", "sentence2": "this may indicate that the subjects are getting fatigued or annoyed (even though their performance is improving).", "label": "neutral", "domain": "Hardware"}
{"id": "test_1672", "sentence1": "We currently know that, in both continu­ous and discrete settings, the only kernel that abides to the scale-space axioms is the heat kernel.", "sentence2": "even though there are indications that the same kernel also applies to graphs [12, 27], a rigorous examination of graph scale-space is -to this point-missing.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1673", "sentence1": "Though Theorem 1 establishes the properties of graph scale-space kernels, it does not provide an explicit form.", "sentence2": "many possible kernels may exist that satisfy the scale-space axioms.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1674", "sentence1": "Considering the high irregularity of wireless networks, columnstochasticity would exacerbate the phantom extrema e.ect.", "sentence2": "with random mean and variance (as compared to seven in our simulations).", "label": "neutral", "domain": "Hardware"}
{"id": "test_1675", "sentence1": "Unfortunately, the gain of the darlington changes with temperature, hence, if its temperature changes, the output current and voltage rises steadily even if the bias is constant.", "sentence2": "for deterministic results, the temperature of the darlington must be kept constant when changing Vbe.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1676", "sentence1": "The power consumption of the board without the use of the LCD screen is 148 mW (24 mA).", "sentence2": "the bottleneck of the present prototype is the 24-bit AtoD which consumes close to 72 mA.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1677", "sentence1": "For monitoring in an apartment, a camera is needed in each room.", "sentence2": "activity recognition using radio signals can cover a much wider area and can see through walls, while cameras cannot.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1678", "sentence1": "We see from Fig. 5(c) that, when RFI is absent, this is probably feasible because walk­ing gives rise to highly fluctuating RSI while non-walking does not.", "sentence2": "in the presence of RFI, the distinction between walking and non-walking is not so conspicuous, as seen in Fig. 5(f).", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1679", "sentence1": "We use these data sets to inves­tigate the effect of bandwidth on location-oriented activity recognition.", "sentence2": "we investigate what happens if we use a bandwidth of 5 MHz, 10 MHz, 15 MHz, 20 MHz, 40 MHz, 80 MHz and 125 MHz.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1680", "sentence1": "Note that, when interference sources exist, all sub-carriers in the secondary data sets are with RFI while only some of the sub-carriers in the primary data sets are with RFI.", "sentence2": "the secondary data sets are worst case scenario with RFI.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1681", "sentence1": "As expected, increasing B gives a better classification accuracy.", "sentence2": "l1-weighting achieves an accuracy of 75%, 95% and 97% when using a bandwidth window B of 5 MHz, 20 MHz and 125 MHz respectively.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1682", "sentence1": "Therefore, we need to consider the RFI environment when training the dictionary.", "sentence2": "the most telling observation is that, for ws = 1, the classification accuracy is merely 57% but if ws = 10 is used, an accuracy of almost 90% can be obtained.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1683", "sentence1": "The wireless devices for RTI can be placed outside the walls to enable them to see through walls or smoke to locate people or objects within the walls.", "sentence2": "rTI has a broad range of applications in emergency response, security (e.g. hostage rescue), health\ncare, and assisted living etc. [4, 9].", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1684", "sentence1": "We propose our networked dynamic sensing configuration solution to achieve robust structural sensing in our system.", "sentence2": "the resolution of the X500 signal is too low to be extracted.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1685", "sentence1": "The area of the single comparator/error detection block responsible to runtime Trojan detection at the final output is also considered in the above QoR function when evaluating its magnitude for both [7] and proposed approach.", "sentence2": "since only a single comparator/error detection block is used in both approaches, hence it has no impact on the QoR of both approaches.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1686", "sentence1": "It is concluded that the appropriate test temperatures for such temperature-dependent defects could be found based on the location of the defects and the properties of the materials involved in those defects [8].", "sentence2": "tests that target these defects are generated.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1687", "sentence1": "It is known that a test's power dissipation depends on the previously applied test.", "sentence2": "the same set of tests when organized differently dissipates different amounts of power.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1688", "sentence1": "In general, hot tests excessively occupy TAM in order to receive heating sequences.", "sentence2": "cold tests underutilize the TAM because of many required cooling intervals.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1689", "sentence1": "Low-Power and Reliable Clock Network Design for Through-Silicon Via (TSV) Based 3D ICs.", "sentence2": "kOZs occupy large layout area thus might  be infeasible in case that layout whitespace is limited.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1690", "sentence1": "When multiple TSVs are present, we assume the TSV and silicon substrate are linearly elastic structures, then according to the stress superposition principle explained in multiple papers [8, ?, ?], the stress coming from several different bodies is the sum of the stress applied separately.", "sentence2": "the stress value at a certain point is the accumulative TSV-induced stress caused by each TSV: .", "label": "entailment", "domain": "Hardware"}
{"id": "test_1691", "sentence1": "Nicolaidis et al. [16] proposed a built-in self-repair method that can be used for on-line faulty TSV recovery.", "sentence2": "in the first experiment, three representative cases of TSV failures, which includes that all TSVs are fault-free, only 1 TSV is faulty, and 2 TSVs are faulty respectively, are considered to verify the function of the proposed scheme.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1692", "sentence1": "Moreover, several types of TSV defects are latent and may easily escape detection during the manufacturing test.", "sentence2": "these latent TSVs are prone to degrade during the field operation and may eventually become faulty and then destroy the entire 3D-stacked IC.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1693", "sentence1": "The yield of 3D DRAM can be significantly elevated since any 2 faulty TSVs within each group can be tolerated through this scheme.", "sentence2": "up to 50% of additional TSVs need to be designed in this scheme, thus rendering a very high cost.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1694", "sentence1": "The logic low value of Pc11 will cause the Pr21 to be set to logic high value and renders disconnecting between Sginal2 and TSV2.", "sentence2": "hence, in order to improve the yield and the quality of 3Dstacked ICs, several faulty TSV repair strategies have been proposed and typically use the neighboring TSVs to replace faulty TSVs.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1695", "sentence1": "Hence, as for more general discussions, we need to discuss not only the case that those assumptions are true, but also the case that those assumptions are not always true.", "sentence2": "in this paper, as the first attempt to establish the framework of temperature dependency aware skew design, we will focus our attention on the former case.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1696", "sentence1": " Additionally, since an application typically processes much more data than the number of instructions executed, most prior cache locking methods, if applied directly to the data cache, would require a large data cache and/or potentially result in runtime overhead in terms of performance and/or energy, since complex runtime analysis wou6 ld be required due to the inherent runtime variability of data caching [21].", "sentence2": "we propose a new methodology for leveraging cache locking for data cache performance and energy consumption optimizations in general purpose embedded systems.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1697", "sentence1": "For example, for gromacs, our methodology locked two phases x and y that comprised of 34% and 17% of the application's execution, respectively.", "sentence2": "assuming this a priori knowledge limits these methods' applicability to general purpose embedded systems (e.g., smartphones, tablets, etc.)", "label": "neutral", "domain": "Hardware"}
{"id": "test_1698", "sentence1": "Yang et al. [21] used a dynamic programming algorithm to determine the locked contents in order to improve the data cache's power consumption and performance.", "sentence2": " Shen et al. [16] showed a strong correlation between data locality and an application's phase characteristics, and showed that data reuse patterns could be used to classify phases.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1699", "sentence1": "Caches are commonly used in embedded systems to bridge the processor-memory performance gap by exploiting the spatial and temporal locality of memory accesses.", "sentence2": "caches can contribute significantly to overall system energy consumption (e.g., the ARM920T's caches consume up to 44% of the microprocessor's overall energy consumption [15]).", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1700", "sentence1": "Much previous work studied cache locking's execution time predictability benefits and phase classification for exploiting an application's runtime variability in isolation.", "sentence2": "little prior work exploits cache locking for optimizing the cache's performance and/or energy consumption while considering runtime variability.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1701", "sentence1": "Such a structure effectively reduces the transistor numbers of a single TCAM cell to 3.", "sentence2": "all these MTJ-based TCAM designs face a severe design challenge the small difference of MTJ s high and low resistance values (usually only a few KO) produces a very limited sensing margin, which significantly degrades the reliability of TCAM design, prolongs the searching latency and induces extra power overhead.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1702", "sentence1": "Compared to SRAM-based TCAM with 12 transistors (12T-SRAM) [9], these designs significantly decrease the number of transistors and therefore reduce the cell area [6][7][10].", "sentence2": "the match line (ML) is connected to a sense amplifier in reading and searching operations, while a pair of source lines (SL and SL) are used to supply the programming or searching data.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1703", "sentence1": "Naturally, the larger κ induces a bigger gap between the sensing voltages of match and miss conditions.", "sentence2": "the voltage margin is larger.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1704", "sentence1": "However, the atomistic treatment of BTI leads to a statistical and con.dence-based perception of those metrics.", "sentence2": "given the variable manifestation of BTI it is imperative to employ statistical methods.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1705", "sentence1": "We observe that the performance is highly dependent on the design topology.", "sentence2": " a significant portion of the datapath corresponds to variation-critical registers, reducing the performance of Algorithm 1 compared to ISCAS85 benchmarks.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1706", "sentence1": "Figure 4 shows that the number of functional samples decreases as we reduce the supply voltage.", "sentence2": "decreasing VDD to meet a power constraint could lead to an increased num­ber of non-functional modules.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1707", "sentence1": "Since the Nangate 45nm library consists of more than 130 standard cells, it is hard to list all the results.", "sentence2": "ten typical standard cells are listed here.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1708", "sentence1": "More importantly, the standard cell library should be carefully designed to enhance the pin accessibility.", "sentence2": "i/O pins need to be balanced distributed within a cell, as the alignment or the densely packing of pins make the cell more difficult to be accessed.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1709", "sentence1": "Besides, the fast method can get very comparable results, in terms of both wire-length and pin access values.", "sentence2": "through fast cell routing the pin access value is 2% less, the wirelength is increased 0.7%.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1710", "sentence1": "Due to the unidirectional shapes of MOL and Metal-1 layers, the patterns are SADP friendly.", "sentence2": "the line-space array decomposition can be applied to SADP with trim masks, with tight control on overlay and wafer-print artifacts.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1711", "sentence1": "The sequence of wire segments is determined based on the LP solutions in descending order.", "sentence2": "the wire segment with the biggest xi is assigned first.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1712", "sentence1": "For compute-intensive applica­tions however, due to the lower number of accesses to the memory, the higher cache access cycles at higher frequencies can be tolerated as it does not affect the IPC significantly.", "sentence2": "for this group of applications the performance increases with the increased frequencies.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1713", "sentence1": "As described in Section 2.2, in order to see the ITD phenomenon, we have to operate the core at lower than nominal voltages.", "sentence2": "for the ITD-aware schemes the operating point of 85% and 75% of the nominal voltage are investigated.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1714", "sentence1": "At high operating voltages, the threshold voltage change will have little effect on (Vgs - Vt) and thus, the mobility will determine the effect of the temperature on the device speed.", "sentence2": "at low operating voltages, the threshold voltage will determine the effect of temperature.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1715", "sentence1": "Assuming 20% of our post-layout top critical path will be RC-delay and 80% cell-delay, the di.erence between pre-layout and post-layout speedup for 25.C to 125.C, will be 7% lower in 20nm at nominal voltage.", "sentence2": "post-layout speedup as a result of increase in temperature is lower than that of pre-layout.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1716", "sentence1": "For R on=10^5 Ω, the critical falling point is around 0.66V.", "sentence2": "the ON and OFF states of memristor can be respectively mapped to HIGH and LOW of V out if setting Vg within the range from 0.34V to 0.66V.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1717", "sentence1": "For spatio-temporal applications, CNNs can offer vastly superior performance and power efficiency when compared to conventional von Neumann architectures.", "sentence2": "there are many ways in which both the performance and imple­mentation complexity of a CNN system could be improved.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1718", "sentence1": "More specifically, our approach dynamically monitors the platform and adaptively adjusts to the COP among multiple cores, using lightweight checkpointing and roll-back mechanisms adapted from Hardware Transactional Memory (HTM) for error recovery.", "sentence2": "we support two distinct types of recovery mechanisms: non-critical and critical.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1719", "sentence1": "Note that the data only need to be logged the first time the address is written within a specific transaction.", "sentence2": "the log size depends on the write footprint of each transaction.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1720", "sentence1": "To showcase the reconfiguration aspects of our proposal with respect to the volume of faults, we present our gathered statistics in two different sets of graphs.", "sentence2": "the top graph in Fig. 6 depicts the simulation results assuming SRAM pfails up to 10-4 (low-pfail case), whilst the graph in the bottom shows the corresponding results for pfails ranging from 5×10-4 to 5×10-3 (high-pfail case). ", "label": "entailment", "domain": "Hardware"}
{"id": "test_1721", "sentence1": "Furthermore, this is not the only advantage of spare mode over VS mode in high probabilities of faults.", "sentence2": "the spare mode has the extra benefit of lower power consumption (in VS mode after a hit at the extra cache a block is exchanged between the main cache and the extra one).", "label": "entailment", "domain": "Hardware"}
{"id": "test_1722", "sentence1": "At the lower level, the LRUL (left) and LRUR (right) show the least recently used subblock in the left/right array respectively.", "sentence2": "the spare mode has the extra benefit of lower power consumption (in VS mode after a hit at the extra cache a block is exchanged between the main cache and the extra one).", "label": "neutral", "domain": "Hardware"}
{"id": "test_1723", "sentence1": "As explained in the previous section, the victim cache, due to faulty block disabling, confronts the misses caused by faults in the main cache as conflict misses.", "sentence2": "the victim cache copes with conflict as well as faulty misses while the spare cache mainly copes with faulty misses.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1724", "sentence1": "After getting the continuous solution, we usually use nearest rounding to project to discrete solution.", "sentence2": "during the nearest rounding, it is hard to tell whether we should round up or round down the continuous size, since it may introduce extra cost when the continuous solution is either rounded up or rounded down.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1725", "sentence1": "This is because it is most suitable for a TCAM-based implementation of the checker which allows storing/parallel-checking against the individual test cubes, containing don t-care bits.", "sentence2": "even after minimization, the number of test cubes can be prohibitively large.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1726", "sentence1": "The FA-Ag procedure results in adding more false alarms per iteration.", "sentence2": "the number of iterations may be smaller because more minimization is possible due to a higher number of expanded cubes per iteration.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1727", "sentence1": "Moreover, unlike the redundant ALU case, the MUXes in (a) are not on the critical path.", "sentence2": "these MUXes can be sized much smaller than the MUXes used for the redundancy technique in Figure 2(b).", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1728", "sentence1": "Finally, column %Imp. represents the performance improvements resulting from the proposed synthesis strategy.", "sentence2": "the gates that have fanout grater than '1' will be duplicated so that their output signal remains intact.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1729", "sentence1": "The BDI method [14] leverages redundant patterns (e.g., zero blocks, narrow values and low dynamic range patterns) exhibited by adjacent data bytes/values within a cache block.", "sentence2": "bDI divides a cache block into several sub-blocks and represents the entire block as two base sub-blocks (a zero sub-block and the first non-zero sub-block) plus the difference information (Δ).", "label": "entailment", "domain": "Hardware"}
{"id": "test_1730", "sentence1": "In contrast, SOCO's compression is dedicated to reducing power consumption in NVMs rather than the traditional goal of saving memory space.", "sentence2": "sOCO leverages a cache-block-level delta-based compression scheme similar to the one used in prior work [14] but utilizes an essentially unmodified page management mechanism to enhance compression effectiveness and reduce over­heads compared with the leading capacity oriented scheme [13].", "label": "entailment", "domain": "Hardware"}
{"id": "test_1731", "sentence1": "We adopt the PCM latency and power model used in prior work [12].", "sentence2": "sOCO s minor degradations are mainly caused by compression/decompression latencies and add a nominal 0.18% and 0.1% for STT-MRAM and PCM, respectively.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1732", "sentence1": "For practical use of polar codes, SCL algorithm that consists of multiple SC component decoders is usually adopted to improve the performance of polar codes.", "sentence2": "the original SCL algorithm is challenged with its inherent long latency.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1733", "sentence1": "Besides, use of LLR instead of likelihood in SCL algorithm was proposed in [6-7] to save area.", "sentence2": "this paper presents LLR-based 2b-SCL decoder that decodes 2 bits simultaneously.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1734", "sentence1": "Word-level compression schemes like FPC reduce bit-writes by leveraging program data statistics to eliminate redundant bit-flips.", "sentence2": "in doing so, the bit-writes for a few of these cells are increased in comparison to other cells within the same word. ", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1735", "sentence1": "For uniform data traffic, the average Hamming distance between the existing and incoming data is the same for mirrored/un-mirrored data.", "sentence2": "fPC and FNW operate only on un-mirrored data; this motivates MNW to take advantage of the possibility to reduce bit-writes using data mirroring, regardless of the success of FNW/FPC.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1736", "sentence1": "Whereas FPC reduces the cumulative bit-writes over uncompressed data with DCW (area under the curve), FPC concentrates peak bit-writes at bit indices 63, 62, and 61, which negatively affects NVM endurance.", "sentence2": "fNW reduces the bit-writes across all bit indices in comparison to FPC, i.e., FNW is better for NVM endurance in comparison to both FPC and DCW.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1737", "sentence1": "In practice, this may lead to premature failure of these cells.", "sentence2": "although FPC is successful in reducing cumulative bit-writes, it may do so at the cost of NVM endurance by increasing the peak bitwrites per cell.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1738", "sentence1": "To overcome these problems, a broad set of solutions that rely on data­encoding [5], write scheduling [6 9], data-migration using address translation [10 12], and architectural improvements [13, 14] have been proposed.", "sentence2": "these techniques address either NVM la­tency, power/energy, or endurance explicitly; this potentially limits their effectiveness in practice.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1739", "sentence1": "The inclusion of an SSL makes the yield of one core in the pair dependent upon the yield of the other: if one core needs the shared spare lane, it is only available if the other core does not.", "sentence2": "there is no straightforward formula to calculate a single core s yield when both SSL and spare cores are present.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1740", "sentence1": "We made the bins in Table 1 as large as possible without complicating their definition (e.g., by combining two disjoint or partially overlapping sets).", "sentence2": "some POF designs do not fit well into the bins.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1741", "sentence1": "A review paper on sub-threshold design was presented in [15], in which the authors suggested an ultra dynamic voltage scaling technique, which dynamically scales the power supply voltage to subthreshold operation.", "sentence2": "traditional dynamic voltage scaling techniques restrict the minimum supply voltage to super-threshold operation [16, 17], providing less aggressive power scaling, while ensuring that PVT variations do not become an issue.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1742", "sentence1": "One possible approach of identifying hotword is to upsample the accelerometer data collected at 200 Hz to 40 KHz, and then reproduce some parts of user's spoken words from the resultant audio file.", "sentence2": "this can incur huge energy cost due to the computational complexity of upsampling as well as analyzing the additionally generated data.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1743", "sentence1": "More recently, in [1], the author employs an anonymous wireless USB device to spy on keystrokes from a wireless keyboard.", "sentence2": "this work relies on the information of the data-transmission protocol of the wireless keyboard and its weak security features.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1744", "sentence1": "As described in the previous section, for a given location of the hacker, the orientation of the hacker relative to the user can change depending on the keystrokes pressed by the user.", "sentence2": "for a given pair of antennas at the hacker, while they might show high sensitivity for a pair of keys pressed by the user, the same pair can demonstrate poor sensitivity for another set of keys at a different orientation.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1745", "sentence1": "The desired sensitivity enables the hacker to distinguish between the location of the user's finger on one key from its adjacent key.", "sentence2": "the vari­ance of this distribution tends to zero as we increase the delay mismatch.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1746", "sentence1": "We observe that the average sensitivity decreases as we increase the delay mismatch.", "sentence2": "while the sensitivity decreases we observe that even for a delay mismatch of one sample duration, the two keys can be reliably distinguished.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1747", "sentence1": "This operation only determines the location of the trough and does not vary the width or robustness of the trough location.", "sentence2": "in this operation we are not changing the delay of the signal but simply changing the phase to ensure location of the trough.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1748", "sentence1": "Measurement noise and multipath can cause varia­tion in the observed phase of the two streams which in turn causes variation in the observed trough location for the same key press in the case of small delay mismatch.", "sentence2": "as we increase the delay mismatch, we decrease the sensitivity of the hacker and the trough location for a given keystroke is more robust to measurement noise.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1749", "sentence1": "For this reason the training information recorded in the experiments is the sensitivity between pairs of antennas for each keystroke.", "sentence2": "while user movements can change the absolute trough locations, the keystroke is still reliably identified by our algorithm since the relative trough location does not change.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1750", "sentence1": "However, when there are changes in the environment, such as a person walking in the vicinity of the user-hacker setup can cause the location of the trough in the spectrum to change.", "sentence2": "it is important for the keystroke detection algorithm to differentiate the cause of movement of the trough location in the cancellation spectrum due to an actual keystroke from that caused due to channel variations.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1751", "sentence1": "With a short ≈ 10 character training input, this accuracy drops to 85%.", "sentence2": "in addition to a trough detection algorithm, the spectrum analyzer block also runs an amplitude matching algorithm.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1752", "sentence1": "As expected from the discussion in Sec. 3.2 the sensitivity decreases as the delay increases.", "sentence2": " the lack of sensitivity assists in reducing the variation of the trough location due to measurement noise.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1753", "sentence1": "Considering the influence of delay mismatch on the performance of the receiver, we observe that for majority of the scenarios considered in this paper, a fixed delay mismatch provides sufficient performance.", "sentence2": "in our implementation, we fix the delay of this partial sample delay block and thus the delay mismatch between the two streams.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1754", "sentence1": "We place the keyboard close to the transmitting antenna and at a lower plane to mimic the setting of a typical laptop.", "sentence2": "further, we illustrated the cancellation performance’s sensitivity.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1755", "sentence1": "It also applies to the quantum case whenever a classically defined oracle arithmetic function must be embodied in the quantum algorithm.", "sentence2": "the first phase (see Fig. 1) of the QDG construction is dedicated exclusively to the forward computations, without taking into account the resetting of the possible ancillae qubits.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1756", "sentence1": "Generally, Adapt should have no worse slowdown than Fix4.8, which is the fastest fixed configuration.", "sentence2": "in a few cases, Adapt is slightly worse.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1757", "sentence1": "For this purpose, COMeT+ keeps a pool of tested free pages to quickly service allocation requests.", "sentence2": "too high of a rate may be harmful: Competition can be caused between testing and the workload.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1758", "sentence1": "The process is inherently memory bound and time consuming, particularly for the most advanced diagnostics that apply several bit patterns in many sweeps.", "sentence2": "most memory diagnostics are done o.ine when the system is not actively serving a workload, such as during boot-up [1] or periods of low utilization (e.g., overnight) [13].", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1759", "sentence1": "The target busy time in phase 2 can be overlapped by different instructions to boost I/O throughput in a single channel.", "sentence2": "it doesn t mean that all instructions can be executed randomly.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1760", "sentence1": "Meanwhile, the multi-level cell (MLC) and triple-level cell(TLC) technologies that constantly reduce the price of flash, make NAND flash based SSDs more competitive than HDDs.", "sentence2": "for better illustration, we have chosen four general workloads for tests, which are Sequential Write, Sequential Read, Random Write, and Random Read.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1761", "sentence1": "Some high-end commercially available PCIe flash storage systems are believed to have used the synchronous interface such as Huawei ES3000 [2], which is alleged to provide up to 770K (4KB block) read IOPS performance and up to 630K (4KB block) write IOPS performance.", "sentence2": "there are few publications on design and detailed implementation of flash controllers based on source synchronous interface.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1762", "sentence1": "Generally, events of cache miss or replacement or garbage collection will trigger FTL to generate flash operation instructions, which are expected to be executed by downstream Dysource controllers.", "sentence2": "the interaction between the FTL and Dysource Controller must strictly follow an interface protocol defined in Dysource Controller.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1763", "sentence1": "As the demands for analyzing more and more data increase, we are faced with new challenges. ", "sentence2": "traditional architectures where data needs to be fetched from memory as to be processed, are becoming a bottleneck.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1764", "sentence1": "In the case of SMG, the 5% maximum performance degradation set for object placement is not aggressive enough to enable the same energy savings as RaPP.", "sentence2": "page migration is necessary to deal with these fluctuations, but page migration comes at a cost: observing characteristics, learning patterns and reading and writing page-sized chunks of memory.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1765", "sentence1": "Reducing the supply voltage of server CPUs is one of the most e.ective methods to lower power consumption.", "sentence2": "the effects of process variations become more pronounced with reduction of supply voltage, leading to failures of many circuits.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1766", "sentence1": "Fortunately, according to [1], the fully parallelized SEC-DED design is simple and fast, with only 14k gates of hardware and one cycle for either the encoding or the decod­ing process.", "sentence2": "the 4EC-5ED design needs about 50k gates and a 14-cycle decoding latency.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1767", "sentence1": "Afterward, our strategy to improve the FTC precision is altered from avoiding fluc­tuations to tracking them.", "sentence2": "a novel version of the FTC, referred to as fuzzy compression adaptive transform (Fuzzy-CAT) is presented.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1768", "sentence1": "As opportunities for future work, we consider setting up predictors for mitigating the effect of long delay inherent in the compression process, as well as automating the optimization of the transform parameters.", "sentence2": "fuzzyCAT exhibits a compelling advantage over the regular F-transform.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1769", "sentence1": "Matrix-vector multiplication in Eq. (1) is the core operation of one of the commonest computation (namely, “recall” process) in NCS.", "sentence2": "cLD is an iterative scheme to update weights (re­sistances) of memristors according to the feedback from outputs [10].", "label": "neutral", "domain": "Hardware"}
{"id": "test_1770", "sentence1": "Unsupervised learning with sparse coding is widely adopted in applications of feature extraction, pattern classification, and compressive sensing.", "sentence2": "even with the state-of-the-art hardware platform of CPUs/GPUs, solving a sparse coding problem is still expensive in computation", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1771", "sentence1": "The block of ETAII (a carry generator and a sum generator) is significantly simpler than those of SCSA and ACAA.", "sentence2": "eTAII has a shorter delay and consumes less power and area than SCSA and ACAA, thus ETAII is selected for circuit comparisons.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1772", "sentence1": "As shown in Fig. 1, an AND gate can implement multiplication: given that the two input stochastic bit streams are independent, the probability of ones in the output bit stream equals the product of the probabilities of ones in the input streams.", "sentence2": "a large c will not help reduce the approximation error.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1773", "sentence1": "To make this error small, c should be large.", "sentence2": "a large c will not help reduce the approximation error.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1774", "sentence1": "We further examined the devices with the largest number of discarded blocks and found that the number of blocks that are discarded is not correlated with the amount of data written to or read from flash cells (examined in Sections 4.1 and 4.2).", "sentence2": "platforms A and B contain SSDs with around two or more years of operation and represent 16.6% of the SSDs examined; Platforms C and D contain SSDs with around one to two years of operation and represent 50.6% of the SSDs examined; and Platforms E and F contain SSDs with around half a year of operation and represent around 22.8% of the SSDs examined.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1775", "sentence1": "We further examined the devices with the largest number of discarded blocks and found that the number of blocks that are discarded is not correlated with the amount of data written to or read from flash cells (examined in Sections 4.1 and 4.2).", "sentence2": "we observe SSDs of both low and high usage across their lifetime (measured by flash cell reads and writes) that have discarded large amounts of blocks and have high failure rates.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1776", "sentence1": "The IEEE 30 bus power system model is used to demonstrate the effectiveness of the simulated testbed.", "sentence2": "with delayed reclosing signal for cyber intrusion, the speed and terminal voltage of the goes out of step and control.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1777", "sentence1": "Smart grid is a complex cyber physical system containing a numerous and variety number of sources, devices, controllers and loads.", "sentence2": "smart grid is vulnerable to the grid related disturbances.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1778", "sentence1": "Simulation of only the designed power system will not incorporate the behavior of SCADA systems.", "sentence2": "simultaneous combination between power systems simulation and SCADA systems simulation is required.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1779", "sentence1": "In this paper, we will demonstrate as they can be used to automatically achieve power management since the earliest stage of the design flow.", "sentence2": "we are focussing on the automation of power gating.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1780", "sentence1": "This is due to the fact that these are the smallest kernels.", "sentence2": "when the smallest kernels are executed and the biggest are powered off, the overhead of the introduced power management logic is less than the benefits of having no switching activity in quite almost of the system.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1781", "sentence1": "A dedicated power controller to properly drive these latter is also required.", "sentence2": "to perform a single-cycle reconfigura­tion, MDC inserts in hardware low overhead switching modules (SBoxes) placed at the crossroads between the different paths of data.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1782", "sentence1": "Moreover, S must also contain a partition of the set of actors Vi/of each input network.", "sentence2": "these issues limit their extensive use to applica­tion specifc scenarios.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1783", "sentence1": "Very often the focus has been on the PG physical implementation optimization, sometimes developing automated flows to cluster and interface different PDs [14]. ", "sentence2": "as far as we know, no automatic frameworks act at a higher level.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1784", "sentence1": "MDC already embeds a power saving automatic feature [12].", "sentence2": "it is able to analyse the input dataflow models and to extract from them the optimal actors partitioning in order to implement clock gating power management strategies.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1785", "sentence1": "Prior research in shared memory for HPC workloads has focused on optimized memory mappings for single OS/R environments.", "sentence2": "in each multi-enclave configuration the performance is more consis­tent than exhibited by the Linux-only environment, demon­strating that our approach does not add variable delays to the runtime of the simulation.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1786", "sentence1": "More general inter-VM message passing tech­niques have been implemented via shared memory [19, 22].", "sentence2": "exascale systems will require broader support for heterogeneous environments than these techniques support.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1787", "sentence1": "While it may be possible to perform shared memory mappings by modifying enclave address spaces remotely, it would likely be difficult to provide such an implementation correctly, and would at least require complicated and inefficient address space synchronization mechanisms.", "sentence2": "first, the co-kernel architecture used to perform cross-enclave message transfers (which are required for the transmission of XEMEM operations) currently restricts all IPI-based communication with the Linux management enclave to core 0 of the system, and thus the presence of multiple enclaves may cause contention for interrupt handlers on this core.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1788", "sentence1": "CTLB performs better than 4 KB pages on small-memory workloads, such as cactusADM, canneal, and omnetpp.", "sentence2": "cTLB provides little benefit on big-memory workloads and performs worse than THP overall.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1789", "sentence1": "This technique optimizes conventional caches by removing the need for tag accesses (and TLB accesses for TCE) on hits.", "sentence2": "but we do measure the network traffic traveling to and from the CPU in order to capture any network traffic variations caused by the stash.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1790", "sentence1": "In our imple­mentation, we require 7 bits of metadata for each run, where runs can be either 2 or 3 bytes long.", "sentence2": "accounting for the metadata, we need only 2 3-byte runs, 4 2-byte runs, or a combination or 2-byte and 3-byte runs in order to free 4 bytes for ECC.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1791", "sentence1": "Finally, Section 5 concludes the paper.", "sentence2": "we find that in the context of COP, a simplified run length encoding can extract the redundancy in the same sign-extended values as FPC with less metadata overhead, allowing more blocks to be compressed.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1792", "sentence1": "To enable these decisions, we keep a simple direct-mapped table of the running average times of different SEED regions.", "sentence2": "similar to dataflow architectures like WaveScalar [33], control dependencies in the original program become data dependencies. ", "label": "neutral", "domain": "Hardware"}
{"id": "test_1793", "sentence1": "Interestingly, well known explicit-dataflow architectures eliminate these overheads by directly executing the data-dependence graph and eschew­ing instruction-precise recoverability.", "sentence2": "even after decades of research, dataflow architectures have yet to come into prominence as a solution.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1794", "sentence1": "In fact, it has been shown that about 85% big data processing tasks can be deferred by a day [40].", "sentence2": "even if the renewable power output is intermittent and time-varying, we can still leverage it for processing many delay-tolerant data sets.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1795", "sentence1": "This is mainly because many data acquisition sites are temporary or difficult to reach -they lack established utility infrastructure.", "sentence2": "standalone power supplies such as solar/wind system (with commodity batteries as energy buffer) are often more suitable for data processing in field.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1796", "sentence1": "Several studies propose perform­ing data analysis while scientific applications producing data [22, 23, 85] or moving computation from compute node to storage servers [86].", "sentence2": "architectural Concerns in Large Datacenters Workshop held in conjunction with the 37th International Symposium on Computer Architecture (ISCA), 2010", "label": "neutral", "domain": "Hardware"}
{"id": "test_1797", "sentence1": "As Figure 2 shows, we explore the opportunity to benefit from data pre-processing using a group of inexpensive, commodity servers that are placed near the data source.", "sentence2": "we are primarily interested in in-situ datasets that need to be processed timely but do not have a very strong requirement for real-time processing.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1798", "sentence1": "The cost benefit of InSURE increases when local data gen­eration rate increases.", "sentence2": "such green energy powered standalone systems are also economical.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1799", "sentence1": "We observe that most of the performance statistics of seismic data are very close to video surveillance.", "sentence2": "this paper takes the first step towards designing server clusters for data processing in the field.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1800", "sentence1": "The optimization effectiveness of InSURE on server-related metrics becomes greater when the solar energy is lower.", "sentence2": "the benefit of our joint spatio-temporal power management actually increases when the standalone in-situ system becomes heavily energy-constrained.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1801", "sentence1": "This overhead will be more pronounced as in-package DRAM size continue to scale up in the future as integration of 4GB [28] and 16GB [3] in-package DRAM will soon be introduced into the market.", "sentence2": "reducing the tag overhead is a primary design goal even for paged-based in-package DRAM caches.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1802", "sentence1": "We would like to thank Jung Ho Ahn and Young Hoon Son for their help with die-stacked DRAM modeling.", "sentence2": "most of prior work along this avenue proposes to augment the memory controller with counters to gather access statistics [12, 30].", "label": "neutral", "domain": "Hardware"}
{"id": "test_1803", "sentence1": "The caching granularity in the proposed tagless cache is aligned to OS page size.", "sentence2": "to effectively increase TLB reach, many modern architectures support superpages (e.g., 2MB, 4MB, and 1GB for x86_64 [9, 18]).", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1804", "sentence1": "Ta­ble 3 summarizes architectural parameters.", "sentence2": "this approach has an ad­vantage of being readily deployable without modifying the software stack.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1805", "sentence1": "In addition, we introduce a simple hardware structure that monitors the locality of data accessed by a PIM-enabled instruction at runtime to adaptively execute the instruction at the host processor (instead of in memory) when the instruction can benefit from large on-chip caches.", "sentence2": "our architecture provides the illusion that PIM operations are executed as if they were host processor instructions.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1806", "sentence1": "Our memory model ensures atomicity between PEIs.", "sentence2": "a PEI that reads from and writes to its target cache block is not interrupted by other PEIs (possibly from other host processors) that access the same cache block.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1807", "sentence1": "To date, most of the existing PIM architectures are based on general-purpose computation units inside memory for flexibility across different applications [14 16, 25, 28, 35, 39 41, 46, 48, 50].", "sentence2": "this introduces two major challenges in seamlessly integrating such architectures into conventional systems in the near term.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1808", "sentence1": "In summary, simple PIM operations interfaced as ISA extensions have great potential for accelerating memory-intensive workloads.", "sentence2": "after the reader PEI execution, if there are no more in-flight reader PEIs to the entry, the entry is marked as writeable.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1809", "sentence1": "First, a local core can continue its execution after invoking a non-blocking remote function call since the core does not have to wait for the termination of the function.", "sentence2": "it allows hiding remote access latency because sender cores can perform their own work while messages are being transferred and processed.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1810", "sentence1": "Adopting a high-bandwidth alternative to DDR3-based main memory based on 3D-stacked DRAM, called Hybrid Memory Cube (HMC) [22], helps this situation to some extent, however, the speedups provided by using HMCs are far below the expected speedup from quadrupling the number of cores.", "sentence2": "once the packet arrives at the remote vault, the network interface stores function arguments to the special registers visible from the core and emits an interrupt for the core.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1811", "sentence1": "For example, in AT.LJ, the partitioning scheme eliminates 53% of non-blocking remote function calls compared to random partitioning (which is our baseline).", "sentence2": "in some workloads, graph partitioning shows only small performance improvement (CT.LJ) or even degrades performance (SP.LJ) over random partitioning.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1812", "sentence1": "The explosion of digital data and the ever-growing need for fast data analysis have made in-memory big-data processing in computer systems increasingly important.", "sentence2": "large-scale graph processing is gaining attention due to its broad applicability from social science to machine learning.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1813", "sentence1": "Adopting a high-bandwidth alternative to DDR3-based main memory based on 3D-stacked DRAM, called Hybrid Memory Cube (HMC) [22], helps this situation to some extent, however, the speedups provided by using HMCs are far below the expected speedup from quadrupling the number of cores.", "sentence2": "if we assume that cores can use the internal memory bandwidth of HMCs2 ideally, i.e., without traversing the off-chip links, we can provide much higher performance by taking advantage of the larger number of cores.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1814", "sentence1": "Unlike host processors that have access to the entire address space of the HMCs, each Tesseract core is restricted to access its own local DRAM partition only.", "sentence2": "a low-cost message passing mechanism is employed for communication between Tesseract cores.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1815", "sentence1": "When there are only a few access ports, the total area of a racetrack memory stripe is mainly determined by the number of domains.", "sentence2": "the area overhead of adding one more read port is moderate (mainly from peripheral circuitry).", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1816", "sentence1": "Qureshi et al proposed adaptive ECC to reduce the overhead of error correction for PCM [27].", "sentence2": "since the shift operation is unique to racetrack memory, detection and correction of position errors are not covered in previous research yet.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1817", "sentence1": "Second, even if ECC detects that misaligned bit is incorrect, it cannot decide the direction and steps of shit errors for certain.", "sentence2": "we have to refresh all data in stripes to refill the correct data (if possible).", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1818", "sentence1": "In other words, the domain is not aligned properly to the access port.", "sentence2": "then, the run-time shift intensity can be calculated as 1/Tinter, which is used to select the safe distance for the current shift operation.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1819", "sentence1": "Such a position error is called \"stop-in-middle\" error in this work.", "sentence2": "the domain is not aligned properly to the access port.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1820", "sentence1": "Unsurprisingly, ShiDianNao significantly outperforms the general purpose architectures and is, on average, 46.38 × faster than the CPU and 28.94× faster than the GPU.", "sentence2": "the GPU cannot take full advantage of its high computational power because the small computational kernels of the visual recognition tasks listed in Table 1 map poorly on its 2,496 hardware threads.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1821", "sentence1": "The spectrum of a pulse train with an arbitrary duty cycle is equivalent via Fourier analysis to a set of sinusoids with various amplitudes at fc and its multiples (harmonics).", "sentence2": "there often are several modulated carrier signals in the same general region of the spectrum, so that their side-band signals may not be neatly separated from each other.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1822", "sentence1": "To avoid clutter, Figure 7 only shows the three parts of the spectrum that contain the left side-bands, the carrier, and the right side-bands of the signals.", "sentence2": "it does not show about 40kHz worth of spectrum to the left and right of the carrier.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1823", "sentence1": "a clock signal becomes stronger or weaker (amplitude-modulated) depending on processor or memory activity.", "sentence2": "modern systems create emanations at thousands of different frequencies, so it is a difficult, error-prone, and time-consuming task to find those few emanations that are AM-modulated by processor/memory activity.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1824", "sentence1": "When a thread is interrupted while waiting at the barrier, the BARRIER instruction would be at the top of the ROB which results in the core sending a SUSPEND request to the MSA tile.", "sentence2": "unlike locks which will simply dequeue the requesting core, for barriers we send FAIL (or ABORT) responses to all participating cores, i.e. we force the barrier to fall back to software.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1825", "sentence1": "Note that one could eliminate OMU entirely by simply allocating/deallocating when an entry is initialized and destroyed.", "sentence2": "this significantly reduces the coverage of the accelerator, since from our evaluation, some applications will use more then thousands of locks.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1826", "sentence1": "The difference is largest in 64-core execution of radiosity, where lock synchronization is frequent, but with many low-contention locks.", "sentence2": "this may be an important consideration for processor manufacturers -after adding the synchronization instructions and our MSA/OMU hardware, the processor man­ufacturer can drop MSA/OMU support in future generations of the processor without breaking compatibility with software that uses the new instructions.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1827", "sentence1": "Moreover, application-specific address mapping schemes that will lead to higher parallelism and locality can be determined via profiling.", "sentence2": "simply changing the address mapping at the runtime will result in incorrect program execution the data need to be reorganized accordingly to retain the original program semantics.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1828", "sentence1": "Section 3.1 discusses how SLIPs are used to determine the insertion and movement of a line in a cache level.", "sentence2": "the latency for each operation (evaluating the energy of all the SLIPs and finding the minimum energy SLIP) was found to be 2 processor cycles at a processor clock rate of 2.4GHz and the throughput is one computation per cycle.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1829", "sentence1": "The for loop body in line 418 rotates the rorig array, and does a streaming access of the array elements from c to r. These array elements are then immediately used in the loop body in line 421.", "sentence2": "the temporal locality of these lines will depend on how close the parameters c and r are to one another.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1830", "sentence1": "The reuse distribution for a sampling page is loaded into the TLB upon each TLB miss, and all lines in that page are inserted with the Default SLIP.", "sentence2": "when a page is in the stable state, the reuse distance distribution is not sampled, and the SLIP stored in the PTE is used to make insertion decisions.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1831", "sentence1": "Both these policies increase energy consumption due to excessive line movements.", "sentence2": "sLIP intelligently inserts lines to an initial location based on the reuse distance history of the line, and avoids excessive line movement.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1832", "sentence1": "NuRAPID and LRU-PEA aggressively promote lines to nearer locations and thereby have a much higher fraction of accesses from those sublevels.", "sentence2": "as we can see from Figure 11, this higher fraction is achieved at the cost of a significant increase in movement energy, which adversely impacts cache energy.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1833", "sentence1": "In a shared multicore LLC, the reuse distance of each reference is higher.", "sentence2": "more references cause misses and end up being completely bypassed from the cache.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1834", "sentence1": "SLIP can be employed in caches using such interconnects to further reduce the energy consumption.", "sentence2": "6b need to be stored per page (3b for L2 and 3b for L3), which can fit in the ignored bits of a 64 bit page table entry (PTE)2 .", "label": "neutral", "domain": "Hardware"}
{"id": "test_1835", "sentence1": "Another 10% of accesses, however, require a larger 256KB cache, and the remaining 24% of accesses do not fit in the cache.", "sentence2": "the system runs Ubuntu 11.04 over Linux 2.6.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1836", "sentence1": "This body of earlier work uses reuse distance to reduce cache misses.", "sentence2": "sLIP employs reuse distance to reduce cache access energy by selecting a set of locations for a line to be inserted or moved into.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1837", "sentence1": "When a thread acquires the global lock for the irrevocable execution of a transaction, all of the other concurrent transactions will be aborted.", "sentence2": "a conflict over program data will abort only some of the transactions that happen to access the same data at the same time.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1838", "sentence1": "A callback requires three messages: {callback, write, data} or {write, callback, data} depending on the relative order between the read and the write.", "sentence2": "fetch&#38;Add to signal one waiting thread.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1839", "sentence1": "The reasoning is the following: in an environment with self-invalidation and self-downgrade, responsibility for coherence is distributed to the core caches when it comes to data-race-free accesses.", "sentence2": "there is no need to track data-race-free data in a centralized directory.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1840", "sentence1": "The evaluation covers a wide variety of parallel applications.", "sentence2": "we evaluate the entire Splash-2 suite [28] with the recommended input parameters.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1841", "sentence1": "Because the divergent applications traverse multiple independent control flow paths, restricting L1 I-cache bandwidth results in a significant performance loss.", "sentence2": "the inclusion of per-slice L0 I-caches, which are probed first when independent warps fetch instructions, vastly decreases the performance loss.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1842", "sentence1": "General Idea of Super Block The notion of super block, first proposed in [25], tries to exploit spatial locality in ORAM.", "sentence2": "it tries to load more than one block from the path in a single ORAM access.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1843", "sentence1": "(a) ocean_contiguous (b) volrend (a) ocean_contiguous (b) volrend Figure 12: Sweep stash size.", "sentence2": "having smaller Z increases the background eviction rate.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1844", "sentence1": "We restrict that only neighbor blocks can be merged into super blocks.", "sentence2": "as in the static super block scheme, only blocks with addresses differing in the last several bits can be merged into super blocks.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1845", "sentence1": "The dynamic super block scheme does not merge blocks during Path ORAM initialization.", "sentence2": " all blocks have sbsize = 1 after initialization. ", "label": "entailment", "domain": "Hardware"}
{"id": "test_1846", "sentence1": "We claim that ORAM with a dynamic super block scheme maintains the same level of security as a normal ORAM.", "sentence2": "adding dynamic super blocks to ORAM does not change the security guarantee of the original ORAM.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1847", "sentence1": "Given any two access sequences A1 and A2 of the same length, ORAM guarantees that the transformed access sequences S1 and S2 are computationally indistinguishable.", "sentence2": "the ORAM physical ac­cess pattern (S) is independent of the logical access pattern (A).", "label": "entailment", "domain": "Hardware"}
{"id": "test_1848", "sentence1": "Z = 4 is chosen here to make it easier to see the performance difference.", "sentence2": "(Note that B' is already in the cache before merging) The changes are written to the Pos-Map block.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1849", "sentence1": "Dynamic mapping methods can overcome the adaptability and compatibility issues that static methods face.", "sentence2": "due to the small instruction scopes and lack of speculation, current dynamic techniques fail to make the best use of routing resources in spatial architectures [10].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1850", "sentence1": "But generally the accelerator does not contain memory ports.", "sentence2": "considering hardware resources is critical when generating mappings for spatial ar­chitectures in order to make effective use of PEs and minimize the datapath usage between them.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1851", "sentence1": "For example, in Figure 2(a), if extra data paths (shown by dotted lines) are applied to allow for forwarding operands from one row to the next, the mapping shown in Figure 2(b) is feasible.", "sentence2": "forwarding requires extra cycles to complete and thus leads to lower performance.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1852", "sentence1": "Ideally, the cache should hold all the pages and all the read operations should be cache hits in DRAM.", "sentence2": "indirections occupy memory, and therefore cache hits in DRAM decrease.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1853", "sentence1": "Second, we are investigating the benefits of combining the memory and file system layers for byte-addressable persistent memories.", "sentence2": " we are evaluating the benefits of a combined indirection layer for leveraging existing file system code as a control plane to manage persistent memory while leveraging virtual memory as a high-performance data plane to access persistent memory.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1854", "sentence1": "Data dominates metadata in our applications which usually map large files to memory.", "sentence2": "having separate translations inside the SSD for metadata of the file system does not add significant overhead.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1855", "sentence1": "POSIX mmap specifies that the OS may pick an address of its choosing to map the file if the address requested by caller is not available.", "sentence2": "as this scheme itself is deterministic, it is still possible for a process to map a file to the same address across reboots.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1856", "sentence1": "MySQL on PCIe SSD's Tpm-C was 8.7x lower compared to MySQL on DRAM for a 480GB TPCC database.", "sentence2": "the SSD setup cost 11.1x less compared to the DRAM setup that makes the SSD setup 1.27x better when performance is normalized by cost.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1857", "sentence1": "Second, we are investigating the benefits of combining the memory and file system layers for byte-addressable persistent memories.", "sentence2": "to address these problems, FlashMap introduces a new virtual mem­ory design where the page table pages needed for mapping a file belong to the file system and are system wide resources shared across processes mapping the same file.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1858", "sentence1": "Moreover, SSDs are non-volatile.", "sentence2": "data-intensive applications are using SSDs as slow, but high-capacity non-volatile memory [39, 35, 38, 5, 45] via memory-mapped files.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1859", "sentence1": "Figure 9 shows the number of networks that would be de­tected.", "sentence2": "one to four random obstructions that cause 1.8dB attenuation per meter of penetration are introduced in the path.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1860", "sentence1": "As said before, our hypothesis is that aerial networks are less affected by building shadowing. ", "sentence2": "we have also performed this measurement in an environment where all buildings are higher than 30m.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1861", "sentence1": "The next bottle­neck will then be the divide for DP (84 cy) and the L3 transfer time for SP (78 cy), respectively, but both of these are purely core-local.", "sentence2": "in column 5 we give the leading dimension below which the respective layer condition is satisfied.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1862", "sentence1": "In the fault not activated cases, the Re­dundant scheme has a significantly greater overhead than the other three schemes, and the overhead increases as n increases.", "sentence2": "for smaller values of n, the normalized runtime of Redundant is less than twice that of the baseline experiments.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1863", "sentence1": "FC-O has the best performance in terms of runtime when no errors are present.", "sentence2": "when errors are present, the overhead of FC-O increases significantly, such that its performance is even worse than the Redundant scheme.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1864", "sentence1": "This is guaranteed by Clover as shown in the proof of Theorem 1 and 2, i.e., all the idempotent regions are resilient against DUE.", "sentence2": "clover can achieve zero DUE as long as program is compiled by Clover.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1865", "sentence1": "That is, even if soft errors have already corrupted architectural states, Clover can recover from the errors, and the detection latency does not need to be zero.", "sentence2": "idempotent processing requires soft errors to be detected within the same region as stated in Section 3.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1866", "sentence1": "In particular, if a region is so small that all its instructions should be protected by DMR (i.e., the tail-DMR frontier is set to the beginning of the region), the sensor may detect an error occurred in the region after it is finished.", "sentence2": "at this moment, the error must have already been corrected based on Theorem 2.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1867", "sentence1": "The error detection latency determines how long the tail part of idempotent regions (tail-DMR) should be to guarantee that its execution time is greater than the detection latency.", "sentence2": "the length of the DMR-enabled part is subject to the error detection latency, i.e., a lower soft error detection latency allows a shorter DMR-enabled part (lower performance overhead) but at the expense of more sensors on the chip (higher area overhead).", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1868", "sentence1": "In partic­ular, Clover avoids such high overhead by integrating idempotent processing that recovers from soft errors by simply re-executing the region in which they occur.", "sentence2": "even if soft errors have already corrupted architectural states, Clover can recover from the errors, and the detection latency does not need to be zero.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1869", "sentence1": "This directly contradicts the assumption, since the DMR detects all the errors that occurs in the tail of the region.", "sentence2": "if the error occurs in the DMRenabled part (i.e., tail) of Rc, the error must be detected by Rc.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1870", "sentence1": "This is necessary to prevent the errors occurring in the tail of a region from escaping to following regions.", "sentence2": "any live-out registers, that are defined in the tail of the region, are required to have check instructions before the re­gion boundary.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1871", "sentence1": "An idempotent region is a part of program code that can be freely re-executed to generate the same output.", "sentence2": "characterizing application memory error vulnerability to optimize datacenter cost via heterogeneous-reliability memory.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1872", "sentence1": "This paper calls such a boundary tail-DMR frontier.", "sentence2": "the compiler determines the frontier so that the DMR-enabled part (i.e., tail) has to be as long as the worst-case sensor-based error detection latency.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1873", "sentence1": "Overall, the full-DMR with with idempotent processing incurs 105% runtime overhead on average.", "sentence2": "clover incurs only 26% runtime overhead on average, which is a 75% reduction, at the expense of only 1% chip area overhead.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1874", "sentence1": "Therefore, the error is to be identified by the sensor-based detector before the tail of Rc finishes.", "sentence2": "it is impossible for the error to escape from Rc.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1875", "sentence1": "[25] uses a classical graph coloring to assign a color to each variable and makes all the variables with the same color share the same space.", "sentence2": "the size of the interference graph is reduced, leading to fewer false dependencies.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1876", "sentence1": "A cost func­tion is proposed for each pre-header of loops to determine whether this pre-header should be selected as a loading point.", "sentence2": "secondly, the false dependency problem is less serious in the iterative ILP approach than in the global ILP approach due to the much smaller size of each y-projection graph of the interference graph, resulting in more eficient utilization of the D-cache.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1877", "sentence1": "The only difference between load and xload is that xload can still load a variable into a cache line even though the cache line is already locked.", "sentence2": "no unlock instruction is needed.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1878", "sentence1": "The attacker is able to physically access the memory when the device is powered off, or at any time during application ex­ecution, and plug it into his/her own device to perform an intrusion attack.", "sentence2": "the attacker does not know how long the program has been executed at the time intrusion is performed.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1879", "sentence1": "Study three will use a design science perspective to retrospectively evaluate current crowdsourcing platform designs to assess the viability of these platforms to support the interests of both services buyers and sellers examined in Studies 1 and 2.", "sentence2": "study 3 will address the following questions: How effective are existing technology crowdsourcing platforms at delivering value to the technology workers and the technology buyers who use them?", "label": "entailment", "domain": "Hardware"}
{"id": "test_1880", "sentence1": "It is posited that the same architectural standards that facilitate distributed intra-firm development, can be used to support crowdsourcing engagements Building on this thesis, Study 2 will explore the awareness, perceived preparation, usage intentions and usage levels of IT crowdsourcing services by IT services buyers.", "sentence2": "study 2 will examine the following research question: What determines enterprise readiness to use technology crowdsourcing services?", "label": "entailment", "domain": "Hardware"}
{"id": "test_1881", "sentence1": "Since the translation is purely combinational, it is also referred to as Combinational Clock Gating.", "sentence2": "only one of these signals can participate in the final gating condition as both these signals are synchronized signals.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1882", "sentence1": "As we can see, LRU method provides relatively low accuracy and high false­alarm rate, which may cause significant backup energy and performance penalty.", "sentence2": "other complicated DBP meth­ods, which are dedicated to high performance applications, introduce unacceptable hardware overheads in an embedded system.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1883", "sentence1": "When combining the above constraints with tight project timelines, it is common practice that designs are taped-out when the ver­ification confidence is deemed sufficient.", "sentence2": "they are the dominant type of design errors that escape to silicon.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1884", "sentence1": "In this work, we explicitly model the uncertain characteristics of system components.", "sentence2": "we describe the reliability of a component using arbitrary reliability functions (exponential, Weibull, custom) with parameters that are arbitrarily distributed between best-and worst-case bounds.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1885", "sentence1": "In our calculations, if an error causes the pipeline to stall persistently or causes an exception in a clock cycle, this exception will happen at every clock cycle from that cycle until the end of the simulation.", "sentence2": "the propagation prob­abilities in these cases are skewed to the cycles closer to the end of simulation.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1886", "sentence1": "The aforementioned protection schemes mainly target scan chains, while protecting user-defined backdoors has gained little attention.", "sentence2": "as chips become more complex and have more debugging functions, these backdoors have introduced more vulnerabilities that need to be addressed [7]-[11].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1887", "sentence1": "Field Programmable Gate Arrays (FPGAs) have some of the advantages of hardware accelerators and are also programmable [3].", "sentence2": "their fine-grain reconfigurability incurs a very high cost in terms of energy efficiency [4].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1888", "sentence1": "The operations from the if-path and else-path form the set of operations Nif and Nelse respectively.", "sentence2": "their fine-grain reconfigurability incurs a very high cost in terms of energy efficiency [4].", "label": "neutral", "domain": "Hardware"}
{"id": "test_1889", "sentence1": "Also, the benefit of having several high-dimensional spaces with a large amount of precise information often yields a computationally difficult problem.", "sentence2": "the remaining parts (motorized stage, EM probe, amplifiers) were made with low-cost off-the-shelf components with an overall price of less than USD 250.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1890", "sentence1": "While SCARE is mainly concerned with recovering individualvalues, e.g., S-box tables, of a (partially) known algorithm,SCANDAL attempts to extract code from side-channel information.", "sentence2": "by measuring and analyzing theside-channel leakage of a µC, the goal is to extract the executed instructions.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1891", "sentence1": "Therefore, they allow designers to explore various architectural options for different design objectives.", "sentence2": " such exploration has exponential complexity, making it practically impossible to explore the entire design space.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1892", "sentence1": "Another reason for the choice of these optimization techniques is that they are commonly available in most of the HLS tools [8].", "sentence2": "our framework can be integrated with a variety of different HLS tools.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1893", "sentence1": "In this work, we address the two main concerns of any digital system design: performance and hardware usage.", "sentence2": "the framework can be applied to include other criterion as well.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1894", "sentence1": "In addition, bank switching results in a more complicated logic for the generation of the addresses to the memory banks, since each bank must be addressed by one of five different combinations of the (i, j) indices as the loop proceeds through the iterations.", "sentence2": "we also need additional multiplexers driving different addresses to each of the four memory banks, depending on the iteration.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1895", "sentence1": "When one task finishes, it will notify all its successors.", "sentence2": "each task should figure out how many predecessors have been finished and how many successors need to be notified.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1896", "sentence1": "However, due to the complexity of accumulated variations, it is hard for designers to determine which TAS strategy works best for a given MPSoC TAS problem.", "sentence2": "quantitative evaluation of TAS strategies is becoming an important issue to guarantee the performance yield in MPSoC design.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1897", "sentence1": "A processed input is delivered to the Code (PIM ) through a finite buffer whose size is defined in the parameter buffer-size.", "sentence2": "there are two cases when the processed input needs to be inserted into the buffer: (1) the buffer has an empty slot, and (2) the buffer is full.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1898", "sentence1": "Typical phylogeny reconstruction methods are based on genomic distance (e.g. neighbor joining), combinatorial optimization (e.g. breakpoint phylogeny), or statistical methods (e.g. maximum likelihood (ML)).", "sentence2": "these platforms typically have mesh-like wired NoCs (obvious choice due to ease of VLSI implementation) and high power budgets.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1899", "sentence1": "The microprocessor power consumption is a major design constraint, which affects performance and reliability.", "sentence2": "it is of decisive importance to co-optimize performance, reliability and power at runtime by applying fine-grained adaptation techniques.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1900", "sentence1": "With more accurate technology data, the inaccuracy would be lower.", "sentence2": "if the self-calibration is employed, the inaccuracy for the combined power consumption is reduced to a negligible value.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1901", "sentence1": "In this regard, the efficiency of the adaptation (performance penalty vs. power/temperature reduction) strongly depends on the available spatial power information.", "sentence2": "the final system performance employing more localized adaption techniques will be considerably better [8].", "label": "entailment", "domain": "Hardware"}
{"id": "test_1902", "sentence1": "Beside the micro-models themselves, it is a challenge to select and frequently access the performance counters that are required to feed the micro-models.", "sentence2": "the access is a very demanding task, as various SPRs (here: 31) need to be accessed at almost the same time (to make sure that the gathered data is correlated).", "label": "entailment", "domain": "Hardware"}
{"id": "test_1903", "sentence1": "Compared to the linear regression based model proposed in [13] that estimates the combined power for each core, our model is much more accurate.", "sentence2": "it can capture the workload trend, which is not possible with simpler models as shown in Fig. 2(a).", "label": "entailment", "domain": "Hardware"}
{"id": "test_1904", "sentence1": "Near Threshold Computing, in which the supply voltage is only slightly higher than the transistor s threshold voltage, is a promising approach to reduce the energy per operation.", "sentence2": "this methodology is highly sensitive to parameter variations and requires a combination of architectural adaptations or specific software techniques to mitigate the effects variability-induced heterogeneity [5].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1905", "sentence1": "In fact, design at the Electronic System Level (ESL)with SystemC as its de-facto standard [7] is established inindustry today.", "sentence2": "the proposed methodology does not only pinpoint the designer to the switch but in addition to that even to the respective parts of the receiver.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1906", "sentence1": "Often, the documentation is not as detailed as needed or became obsolete due to changes in the implementation that have not entirely been propagated [10].", "sentence2": "feature localization is a crucial step within the design process which, thus far, has mainly been conducted manually (e.g. by inspecting the HDL implementation).", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1907", "sentence1": "The features a designer may look for in this system may be concerning the distribution of data.", "sentence2": "which parts of the design are triggered when a package is sent to a specific destination might be of interest.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1908", "sentence1": "The dynamic behavior supported by SystemC particularly differentiating between multiple instantiations of the same type of class are not supported.", "sentence2": "features might not be trackable.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1909", "sentence1": "Cell voltage increases monotonously with SoC, however.", "sentence2": "encouraged by these results, we are interested in the design of components specifically for the purpose of cell balancing.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1910", "sentence1": "To exploit reuse, the data in the window buffer (stage 1b) is shifted left and the next two columns of data are loaded.", "sentence2": "this also means that the programmer has to have knowledge of hardware design and perform the optimization steps (e.g. extracting parallelism) manually.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1911", "sentence1": "Current design methods for accelerator synthesis, such as High-Level Synthesis, are not fully automated.", "sentence2": "this paper investigates the use of this algorithmic skeleton methodology for FPGAs.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1912", "sentence1": "Note that OCL formally considers the right-hand-side of an equation as the argumentof the = -operation called on the left-hand-side.", "sentence2": "the model element songs would be classified variable if both sides of the equation were swapped.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1913", "sentence1": "At the same time, we aim for not leaving the burden of generating frame conditions entirely on the designer.", "sentence2": "a systematic methodology for the generation of frame conditions is proposed which assists the designer in a comprehensible fashion and with automatic methods in this process.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1914", "sentence1": "In summary, an SCC being combinational is equivalent to every loop in the SCC being dominated by one controlling side-input.", "sentence2": "if there exists one loop whose sideinputs are all non-controlling values for an input assignment, the SCC is non-combinational.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1915", "sentence1": "Furthermore, if OSIs are within the fanout cones of other SCCs, they also possibly have indefinite values.", "sentence2": "only OSIs that do not have loops in their fanin cones are considered for self-conflict detection.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1916", "sentence1": "Once all SCCs in a circuit pass the validation procedure, the combinationality of the circuit is guaranteed.", "sentence2": "in the following discussion, for simplicity, we only focus on the SCC which represents a cyclic sub-circuit of the original circuit.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1917", "sentence1": "Mobile applications often heavily use shared libraries and system codes, leading to a large instruction working set and high instruction L1 cache miss rate [7] [8] [3].", "sentence2": "the L1 data miss rate keeps relatively low because the data access patterns in mobile applications exhibit good locality in the L1 cache [7] [8].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1918", "sentence1": "Moreover, the access latency to the CIHT table is also negligible comparing with the long latency on an L2 miss.", "sentence2": "our power model is built on top of McPat [10] and the STT-RAM is modeled by the NVSim [5].", "label": "neutral", "domain": "Hardware"}
{"id": "test_1919", "sentence1": "Mobile devices are generally powered by a finite-capacity energy source - a battery, whose size and capacity are quite limited due to the restricted physical size of the mobile devices.", "sentence2": "energy consumption is undoubtedly the most important factor in modern mobile system design, and the energy-efficient computing (i.e., achieving substantial energy savings without degrading the performance) becomes one of the key challenges faced by the computer architects and system designers. ", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1920", "sentence1": "However, the write operation of the STTRAM, takes longer time and consumes more energy.", "sentence2": "all these motivate us to separate the user and kernel blocks in the L2 cache to preserve the useful blocks.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1921", "sentence1": "Moreover, the access latency to the CIHT table is also negligible comparing with the long latency on an L2 miss.", "sentence2": "there is almost no performance overhead in our technique.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1922", "sentence1": "Aggressive technology scaling has allowed the continuous increase of processor performance while maintaining the same power consumption across generations.", "sentence2": "the worsening of variations and the enormous design margins enforced by conventional techniques for avoiding timing failures limit the performance returns from technology scaling and lead to power overheads especially in sub-40 nm nodes.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1923", "sentence1": "Particularly one type of highly pessimistic margin has already burdened the traditional design flow even long before the recent rise of variability issues: the traditional synchronous design paradigm determines the operating frequency according to the worst critical path that is identified through static timing analysis under assumed worst case conditions.", "sentence2": "this critical path is not always excited, thus leading to processors that perform (much) slower than what they can theoretically achieve.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1924", "sentence1": "In conventional many-core processors that dont have dark silicon constraints, there exists only one TDP mode, i.e., powering-on all cores at nominal voltage/frequency to maximizing performance within a TDP constraint.", "sentence2": "dark silicon many-core processors exhibit a multitude of TDP modes, each resulting in a starkly distinct thermal and leakage power profile.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1925", "sentence1": "[1,H]) denote the coordinates of the core i.", "sentence2": "dark silicon many-core processors exhibit a multitude of TDP modes, each resulting in a starkly distinct thermal and leakage power profile.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1926", "sentence1": "In the nanometer era, the exponential dependence of leakage power consumption on threshold voltage has constrained further threshold-and supply-voltage scaling.", "sentence2": "the power density is increasing with technology scaling at a rate that it cannot be sustained.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1927", "sentence1": "Note that if we consider only two reconfiguration points (e.g., 64/2 and 32/4), the memory overheads are 6.9%.", "sentence2": "for a fair comparison, the overheads of DARCA must be compared against prior redundancy-based FT schemes.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1928", "sentence1": "This is because at the specific pfail , DARCA is configured to operate as 16/8 cache (inflection point between 32/4 and 16/8 configurations).", "sentence2": "the 16/8 configuration depicts a strong prefetch-friendlybehavior.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1929", "sentence1": "Thereby, the storage requirements between DARCA and a typical 32-entries DBR scheme are comparable.", "sentence2": "dARCA follows a different approach. The extra storage area is devoted to hold information to controlhow the cache data array disambiguates the cached elements(control-based redundancy).", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1930", "sentence1": "Moreover, when prefetching in employed (bottom graph of Fig. 5), DARCA manages to further increase its distance from prior FT schemes.", "sentence2": "when the block disabling technique is employed, the baseline cache will be able to keep data in cachelines even if they have some faulty subblocks i.e., four faulty­bits are attached in each 64-bytes physical cache frame or onefaulty-bit every 16-bytes [1].", "label": "neutral", "domain": "Hardware"}
{"id": "test_1931", "sentence1": "This is important since ourtarget is to increase the reliability of the latency-sensitive L1s.In order to calculate access time, we modified CACTI source code to integrate the DARCA extra components into the design.", "sentence2": "the extra components are i) the variabletag selection logic (below tag arrays), ii) the index bit selectionlogic (before cache decoder), and iii) the extra level ofmultiplexer (below data arrays).", "label": "entailment", "domain": "Hardware"}
{"id": "test_1932", "sentence1": "The thermal stability coefficient (ǻ) is dependent on the volume of the free layer and on the temperature of the device.", "sentence2": "maintaining the same area (A), the thermal stability can be modified by modifying the thickness of the free layer.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1933", "sentence1": "The TMR ratio is therefore defined as: TMR= (RH RL)/RL.", "sentence2": "maintaining the same area (A), the thermal stability can be modified by modifying the thickness of the free layer.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1934", "sentence1": "It is integrated in our advanced SystemC-TLM2.0 virtual-platform setup [21].", "sentence2": "it can also be integrated in other simulation environments like gem5.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1935", "sentence1": "The coverage of the test with data pattern 0xFF alone is relatively low compared to the WIDE I/O DRAM results (39-43% versus 83% for WIDE I/O).", "sentence2": "we focus on dswG ... channel 3 only to show retention time depending bit .ips.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1936", "sentence1": "On the other hand, in SUOR ONoC, the worst-case SNR quickly decreases with an increasing MR passing loss.", "sentence2": "the worst-case SNR is above 16dB when the passing loss is -0.001dB, but it is only above 2dB when the passing loss is -0.01dB.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1937", "sentence1": "Regarding coherent crosstalk, it is generated by the switching MRs of cluster 0 (the sending cluster).", "sentence2": "it should be noted that there exists a case where one cluster finishes its communication on almost all clusters and injects those tokens into the waveguide.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1938", "sentence1": "We can observe that the impact of variations on the mean and variance of the delay for different configurations is shown in Figure 6 and Figure 7.", "sentence2": "the mean delays of the three configurations are similar, while the spreads around the mean differ substantially.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1939", "sentence1": "In absence of the final clause, T4 would not necessarily execute immediately, and so the timing behaviour of T1 would change as well.", "sentence2": "task T0 waits on the taskwait at line 16 until tasks T1 (not its descendant task T4) and T2 have completed before proceeding past the taskwait.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1940", "sentence1": "V denotes a sequential operation or job, characterised by a worst-case execution time (WCET) estimation.", "sentence2": "the WCET estimation of T0 depends on the timing behaviour of T1,T2,T3 and T4.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1941", "sentence1": "This confidence level expresses how sure the players are about the real value of the varying parameter.", "sentence2": "the player's confidence determines the magnitude of the variation.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1942", "sentence1": "Solving such large systems at frame rates of up to 30 fps is computationally very demanding and often a pixel dense solution is not required.", "sentence2": "the problems are usually solved on around 10× sub-sampled grids in order to reduce the computational complexity. ", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1943", "sentence1": " Moreover, due to Moore’s law, ICs will be more and more densely integrated, and so this method will become more and more difficult in the next future. ", "sentence2": "there is a need for medium cost HT optical detection methods.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1944", "sentence1": "Standard destructive reverse-­engineering techniques (usage of Chemical Metal Polishing followed by Scanning Electron Microscope image reconstruction and analysis) can be used to detect HTs if we can compare ICs with a golden circuit.", "sentence2": "this task is very expensive since it takes a lot of time to realize properly: if the obtained images are blurry for any reason (like bad mechanical or chemical preparation), the reverse-engineering is hard.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1945", "sentence1": "If the HT is close to a net Na (or directly connected to a net Na), dHTa is defined as the random delay added by the HT to the net Na in an infected IC.", "sentence2": "impact of two HTs on delays, for (P, K)no13 (up) and no47 (down) AES bits, estimated by the number decrements by clock steps One can highlight on the importance to study not only the needed to fault each considered bit.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1946", "sentence1": "Using this model, the probability of detecting a HT can be calculated.", "sentence2": "we can estimate the false positive and false negative probability of HT detection as a function of HT size for a set of ICs.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1947", "sentence1": "The manufacturing flow is not impacted by this detection method.", "sentence2": "each client could decide to use or not this detection method depending on his security needs.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1948", "sentence1": "As a result any promising solutions that require high ratio of 1-to-M and/or large multi­cast destination groups, are avoided [5], [6].", "sentence2": "there is a need to eliminate these constraints and improve performance by proposing a 1-to-M interconnect architecture.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1949", "sentence1": "This paper proposes a novel hybrid architecture, which improves the on-chip communication bandwidth significantly using mixed wires and surface wave interconnects (SWI) fabrics.", "sentence2": "the bandwidth of multicasting can be drastically improved.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1950", "sentence1": "This is especially true for chip multiprocessors (CMPs) that were introduced to provide near­linear performance improvements when complexity increased (Pollack s rule), while maintaining lower power and frequency budget [4].", "sentence2": "the evaluation results show significant improvements in terms of average delay and power consumption with a relatively small die area penalty compared to other architectures.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1951", "sentence1": "In addition, this type of interconnect has the cheapest implementation cost compared to other fabrics.", "sentence2": "the best solution would be to combine both interconnects, metal and SW, in one hybrid wire-SW interconnects in a multi-layered network architecture; in short W-SWI, as shown in Fig. 3.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1952", "sentence1": "Low-cost and efficient routing solutions based on Dimension-Ordered Routing, like FDOR [9], or on turn models [10], [11] have been proposed to provide coverage on irregular topologies.", "sentence2": "they either provide low coverage (single link faults) [9], [10] or disable a large number of faultfree nodes to extend the coverage [10], [11].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1953", "sentence1": "The technical scaling in the nano-era has enabled smaller, faster, and low-power transistors that provide high performance-per-power effi­ciency.", "sentence2": "ultra-small dimensions and low operating-/threshold voltages have led to various reliability threats like soft errors and aging [1, 2].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1954", "sentence1": "So the speedup of ACSEM on quadsub is quite low, since the step value is very close to or even larger than the average length of paths in the circuit graph.", "sentence2": "the BDFS algorithm with the step mechanism will be degraded to a full BDFS algorithm.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1955", "sentence1": "Ignorance of the circuit s error masking property may lead to over-design of a reliable processor or application-specific hardware accelerator circuit that may incur signi.cant power and area overhead2 .", "sentence2": "a fast analysis of the error masking characteristics is required to evaluate and enhance the quality of a reliable processor design in a cost-effective way, while curtailing the analysis and design time to ensure fast time to market.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1956", "sentence1": "Thus, unlike earlier comparisons made in tables II, III and IV, we cannot give a performance guarantee for I-ordering + DP-fill over techniques proposed in [20], [21] or [22].", "sentence2": "it is interesting to note from Table V that the proposed I-ordering + DP-fill technique actually outperforms all the these techniques for most of the benchmarks and the percentage improvement increases with increase in circuit size.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1957", "sentence1": "The new model retain more information of the circuit, yet it can still be efficiently handled by existing graph partitioning methods.", "sentence2": "it is expected to provide a better partitioning compared to traditional S-graphs.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1958", "sentence1": "The TSV which is located at the corner is connected a 3-to-1 MUX whose inputs are its own signal, signal from the same ring and signal from the outer ring.", "sentence2": "the TSV which is not located at the corner is connected to a 4-to-1 MUX whose inputs are its own signal, the signal from the same ring, the signal from inner ring and the signal from outer ring.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1959", "sentence1": "In Figure 4, ãrst, the signals of TSVs in the ãrst ring can be shifted to their neighboring TSVs in any direction.", "sentence2": " these signals can be shifted to neighboring TSVs in the ãrst ring clockwise and counterclockwise, and to neighboring TSVs in the second ring.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1960", "sentence1": "As W or S shrinks, the read latency becomes larger due to larger wire resistance (inset).", "sentence2": "we investigate the effect of synaptic device’s on-state resistance (RON) that stands for the maximum D value.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1961", "sentence1": "Each individual liquid crystal can be tuned by changing the electric potential applied to it in order to let more or less light pass.", "sentence2": "a key feature of an LCD display design is the presence of a source of light to illuminate the array of liquid crystal cells from behind.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1962", "sentence1": "When analyzing this pirate video sequence, each screen is successively tested as a potential pirate device.", "sentence2": "screen 1 and Screen 3 are expected to have an aliased temporal fre­quency ft close to 10 Hz with the Panasonic camcorder.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1963", "sentence1": "However, these benefits are obtained at the cost of significantly higher arithmetic effort: Due to the comparative complex underlying mathe­matical method, a large set of additions and multiplications is required1 .", "sentence2": "the algorithmic performance evaluation highlights our work as very efficient in terms of arithmetic calculation effort (see Tab. 1).", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1964", "sentence1": "Unless a thermal violation is predicted, the decisions of the default drivers such as the core and GPU frequencies, choice of big or little cluster and number of active cores, are affirmed.", "sentence2": "the proposed DTPM approach is non-intrusive when the temperature is within permissible levels.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1965", "sentence1": "We emphasize that using a fan is not feasible when this chip is used in a smartphone or tablet, which is the case for Samsung Galaxy S4.", "sentence2": "in general, this analysis is repeated for each computing resource such as the little cores and GPU to extract the corresponding leakage power model.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1966", "sentence1": "Constraining the maximum frequency to limit the temperature, while passively waiting for thermal violations and reacting by throttling the cores, impairs performance as well as reliability by causing large temperature variations [7].", "sentence2": "predictive ap­proaches can take advantage of rich set of dynamic configuration capabilities to manage temperature effectively [8].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1967", "sentence1": "Thermal modeling and dynamic thermal management have received significant attention due to increased power densities and reliability implications of temperature.", "sentence2": "poor performance of reactive approaches led researchers to develop compact thermal models [7, 9] and thermal prediction techniques [8, 10, 11] which can be used in pro-active thermal management methods.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1968", "sentence1": "Thorough experimental evaluation shows that the proposed approach not only eliminates the need for a fan, which is not a viable choice for mobile devices, but also provides significant power, thermal, and reliability advantages.", "sentence2": "it regulates the temperature more effectively than the default configuration which uses a fan, and on average offers 10% platform power savings with 3.3% loss in performance.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1969", "sentence1": "We ran each benchmark and predicted the temperature T [k+10] at every control interval T [k].", "sentence2": "the temperature one second (10 control intervals) ahead of time is predicted using a sliding window, and the predictions are compared to the measured values at the end of each experiment.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1970", "sentence1": "Furthermore, traditional NoC router uses virtual channel allocator, switch allocator, arbiter and a crossbar to calculate and allocate flits to the relevant output.", "sentence2": "in our design, we just use direction channel allocator and three separated arbiter to realize same function.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1971", "sentence1": "Both of them always grant real-time packets higher priority over best-effort packets and employ round-robin arbitration between packets of the same traffic priority class.", "sentence2": "rigid round-robin is modified by giving the most recently served port the lowest priority, which increases the fairness of the arbitration as illustrated in Fig. 4.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1972", "sentence1": "To achieve such a writing scheme, a multi-step writing should be applied in GSHE logic device writing.", "sentence2": "since writing of these device depends on the direction of charging current, once a device has been written, it cannot perform the same function again.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1973", "sentence1": "As load resistance of writing devices is reduced by parallel structure, the current is mainly determined by MTJ resistance.", "sentence2": "such a structure will distribute charging current through each strip, the current will largely reduced depends on how many output elements there is.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1974", "sentence1": "However, with a fixed RM stripe length, overlapping more cells reduces the space for each cell to allocate their access ports.", "sentence2": "without expanding the overlapped cell area, fewer ports per racetrack stripe leads to longer shift distance.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1975", "sentence1": "Meanwhile, the generation of magnetic field will cause additional energy consumption.", "sentence2": "the area changes from 2.66mmto 5.03mm, by changing the optimization target from area to shift latency.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1976", "sentence1": "We consider the four learning procedures listed in Table I.", "sentence2": "the BMF learning procedure is compared to the conventional learning procedure that uses as training set (a) only the late-stage real inliers, (b) a raw combination of early-stage simulation data and late-stage real inliers, and (c) only the early-stage simulation data.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1977", "sentence1": "The non-intrusive sensors consist of dummy analog stages and single components extracted directly from the topology of the LNA.", "sentence2": "we consider a dummy bias stage identical to the bias stage of the LNA formed by transistor M3 and resistor R1, a dummy gain stage identical to the gain stage of the LNA formed by transistors M1 and M2, a dummy diode-connected transistor identical to the transistor M1, and a dummy capacitor identical to the capacitor Cin.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1978", "sentence1": "The acceleration of applications, running on a general purpose processor (GPP), by mapping parts of their execution to reconfigurable hardware is an approach which does not involve program's source code and still ensures program porta­bility over different target reconfigurable fabrics.", "sentence2": "the problem is very challenging, as suitable sequences of GPP in­structions need to be translated/mapped to hardware, possibly at runtime.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1979", "sentence1": "Finally, the communication is send over a PCIe link using a kernel mode driver.", "sentence2": "all communication between the host and the kernels goes through this layers and can cause a significant overhead.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1980", "sentence1": "Modern heterogeneous devices contain tightly coupled CPU and FPGA logic, allowing low latency access to accelerators.", "sentence2": "designers of the system need to treat accelerated functions specially, with device specific code for instantiating, configuring, and executing accelerators.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1981", "sentence1": "DMSD needs, however, periodic measurements of end-to-end packet delays rather than of injection rate.", "sentence2": "the receiving nodes, and not the transmitting ones, measure the delay.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1982", "sentence1": "A very relevant application in this domain pertains to the development of the gene therapy for HIV-1 [11], which has to tackle the dynamic variation in parameters of the biological interactions when searching for the best configuration and amount of engineered cells.", "sentence2": "we explore the computation kernel that studies the interactions between cells and virions and enables the design of a gene therapy for HIV-1 infection, and consider it as a use-case to design and evaluate the potentials of a NoC-based multicore platform.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_1983", "sentence1": "In the actual system, each region has a fluid boundary.", "sentence2": "in our abstraction, we model these regions as cubes in three-dimensional space.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1984", "sentence1": "In this paper, we present a novel performance simulation tool with the added flexibility adequate for use in architectural optimization as the parameters of the component devices change.", "sentence2": "this simulator is designed to handle reconfigurable quantum hardware, where device parameter (DP) values can be adjusted  handle reconfigurable quantum architecture, where re­source investment scenarios can be studied  output concrete breakdown of performance metrics (e.g. total execution time and failure probability) and contents of resource overhead (such as qubits used as ancilla or for communication purposes).", "label": "entailment", "domain": "Hardware"}
{"id": "test_1985", "sentence1": "The fidelity degradation of a BSC as a function of distance tends to limit its utility to short distance communication only.", "sentence2": "eL can be realized with distance-independent fidelity, which makes it a suitable candidate for long distance communication.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1986", "sentence1": "However, VOSPA always outperforms OPT as long as N is larger than 2.", "sentence2": "vOSPA achieves a reactant reduction by 22% and 37% as compared to OPT if a Mixer-4 and Mixer-8 is utilized, respectively.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1987", "sentence1": "For DMFBs under the (1:1) mixing model, one of the two input solutions must be a 1/2 1/2 buffer solution.", "sentence2": "the resultant CV is always half of the 1/2 original CV after an exponential dilution operation.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1988", "sentence1": "A dozen of approaches have been proposed to address this problem in past few years.", "sentence2": "to the best of our knowledge, there is still no algorithm aiming at the problem on FMFBs equipped with multi-segment mixers.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1989", "sentence1": "It relies on optical waveguides to carry optical signals, so as to replace electrical interconnect, and provide the low latency and high bandwidth characteristic of optical interconnect.", "sentence2": "the temperature gradient among ONIs influences the SNR, as detailed in the following.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1990", "sentence1": "Then, MR heater power is explored through thermal simulations in order to reduce the gradient temperature.", "sentence2": "the active layers are surrounded by Si/SiO2 line constituting the mirror structure, which allows coupling the light into the horizontal waveguide (optical signal shown in blue arrow) using a tapper with an estimated 70% efficiency.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1991", "sentence1": "For the sake of simplicity, we assume unrestricted and immediate access to all potentially shared resources such as a bus and only concentrate on the execution cycles of the target application in isolation.", "sentence2": "our framework still improves over the naive approach.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1992", "sentence1": "In the available system, cores 0-7 belong to processor with ID 1, whereas cores 8-15 belong to the processor with ID 2.", "sentence2": "thereafter, a model is estimated from the given calibration trace and the observed temperature trace (line 21), see [12].", "label": "neutral", "domain": "Hardware"}
{"id": "test_1993", "sentence1": "A model is separately computed for each discrete frequency f ∈ F since the thermal properties of the system may change considerably with frequency.", "sentence2": "it may not be possible to simply scale a given model derived at one frequency and use it at a different frequency.", "label": "entailment", "domain": "Hardware"}
{"id": "test_1994", "sentence1": "One common approach is to use heat maps captured from the silicon layer of the given processor as it executes instructions, and use heuristic algorithms to back-estimate the detailed power model to be used with the Hotspot simulator, see [8].", "sentence2": "several critical details (e.g., processor floorplan, on-chip power and/or temperature management algorithms) must still be guessed.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1995", "sentence1": "This process is not always conscious, therefore making it difficult to understand the process.", "sentence2": "once they were selected, starting ideas are characa­terized as either similar to previous designs (analogies [11]), combinations of existing design features [12], generalizations, e.g., through induction [7], and sudden insight (like through restructuring the knowledge representation).", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1996", "sentence1": "The first two steps of the reasoning-based synthesis flow are the same as for the flow in Figure 3(a).", "sentence2": "c. Causal parts present how the features of two or more concepts are combined to form a new concept that offers improved performance and/or leads to a less constrained solution.", "label": "neutral", "domain": "Hardware"}
{"id": "test_1997", "sentence1": "As expected, the small delay fault coverage is generally lower than the transition fault coverage.", "sentence2": "there are also numerable cases of small delay faults, which are detected despite their corresponding transition faults being not detected.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1998", "sentence1": "The host system contains 16 IntelR XeonR processors clocked at 3.4GHz and 256GB of RAM. ", "sentence2": "the peak memory consumption on the host system never exceeded 10GB.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_1999", "sentence1": "This section describes the proposed algorithm for functional ATPG.", "sentence2": "this means that in practice the generated test programs require a manual step to check their compliance with the SBST constraints, triggering a complex and ex­pensive transformation in the likely case that they don t match them.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2000", "sentence1": "The measurements collected from attached external thermistors were used as the expected value for the skin temperature and screen temperature.", "sentence2": "we have used the CPU temperature, utilization, and frequency, as well as battery temperature collected in our log file.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2001", "sentence1": "If the yield value is less than the minimum required yield, the solution is eliminated and the second phase of yield analysis is not performed.", "sentence2": "the synthesis time is reduced since many redundant simulations are not run.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2002", "sentence1": "For instance, slightly losing from one electrical constraint may cause much better performances for the other constraints, as a result enhances the overall circuit performance.", "sentence2": "the design coefficients are added to the developed tool in attempt to arrange tradeoffs between each design constraints.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2003", "sentence1": "We logicaly divide our multi-core chip into memory islands based on the number of memory controllers.", "sentence2": "at each node we can have maximum of 2 different routers that are optimized for 2 VF pairs (e.g, node i and node i +1 in Figure 2).", "label": "neutral", "domain": "Hardware"}
{"id": "test_2004", "sentence1": "For what concerns dynamic metrics estimation, in a traditional testbench based on transient simulation, the converter samples a sinusoid according to Shannon's law.", "sentence2": "the mismatch and the parasitics of each capacitance affect the conversion accuracy.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2005", "sentence1": "Ldigital should contain all the functional blocks, that form the power-down part.", "sentence2": "structures like e.g., inverters, nand and nor gates, level shifters and pass-gates need to be detected.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2006", "sentence1": "The aging effects can be investigated using aging simulators [10].", "sentence2": "approaches based on numerical simulations are computationally expensive and, therefore, infeasible for large designs.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2007", "sentence1": "The breakpoint-based silicon debug approach allows users to stop the normal (system) operations of the circuits under debug (CUDs), extract the internal states of the CUDs for examination, and then resume the normal operations for further debugging.", "sentence2": "then when the CUD_MA is granted to use the bus, it will inform the PAM_IP_BUS to unlock the system and to accept data from the CUD.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2008", "sentence1": "If the CUD is interrupted during a transaction in which it serves as a transmitter, the receiver will never get the correct data in a valid transaction because the transaction is not replicable.", "sentence2": "such transactions should not be interrupted.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2009", "sentence1": "The weight of the remaining cells are calculated base on the modified error response.", "sentence2": "in recent years, advancement of convolutional compactors [4], [5] which are based on a finiteinput-response principle, have added a new dimension to scan response compaction.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2010", "sentence1": "In the swap disabled scheme, applications may get terminated by the LMK when there is no sufficient memory.", "sentence2": "switching between those applications could take longer time as they will have to be reloaded from flash storage.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2011", "sentence1": "Since NVRAM is byte-addressable, all the swapped out pages can be accessed directly through load and store instructions.", "sentence2": "one does not have to swap in pages for read requests.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2012", "sentence1": "However, we found that page type is usually available to the OS, and it is easy for the OS to differentiate code and data pages.", "sentence2": "we argue that a better solution is to swap out code pages to the swap area and execute the code in-place, without copying them back to main memory.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2013", "sentence1": "As indicated in Figure 1 in decode, read, and write-back stages all instructions share the same logic and therefore the specific pipeline stage delay is not instruction-specific.", "sentence2": "in the execute stages each instruction has separate dedicated logic, each with its own delay.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2014", "sentence1": "Consequently, even though some instructions can operate at high frequencies, the slow instructions determine the processor clock period, resulting in the underutil­isation of the processor potential.", "sentence2": "the fast instructions latent performance may be harnessed through Adaptive Clock Management (ACM), i.e., by dynamically adapting the clock frequency such that each instruction gets sufficient time for correct completion.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2015", "sentence1": "Incorporating PCM into FPGA designs has been attempted previously in a rather simplistic fashion; by replacing SRAM cells with 1T2R PCM cells [3].", "sentence2": "we explore the use of PCM to exist in multiple states and thus store multiple bits per cell.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2016", "sentence1": "However, the exponential rise in delay due to the parasitic capacitance becomes painfully evident as the size of the LUT increases.", "sentence2": "lUT size of 12 is over five times slower than it s SRAM counterpart.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2017", "sentence1": "Whilst operating in the single-output mode, the LUT utilizes an additional input signal to select either the most significant bit or the least significant bit as the output (Out A).", "sentence2": "the above described LUT can operate either as a K-input LUT with a single output, or a single LUT with K-1 inputs with two outputs.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2018", "sentence1": "This results in a small area overhead when compared to a traditional SRAM-based LUT.", "sentence2": "since every LUT memory cell can share all the sense amplifiers, the area overhead is diminished.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2019", "sentence1": "However, since the resistance of each of these states is higher than that of the SET state, the read speed for the cases 01 and 10 is higher than that of 00 and 11.", "sentence2": "on an average, the overall read speed has been sacrificed.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2020", "sentence1": "However, the larger the memory cell, the longer the interconnect.", "sentence2": " if we reduce the area overhead of the configuration memory, we can reduce the delay through the crossbar as well as the overall size of the CLB which in turn would reduce global routing overheads.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2021", "sentence1": "For an accurate model, we scaled the resistance, capacitance and delay parameters specified from 40nm to 22nm.", "sentence2": "resistance and capacitance values for routing channels and devices were scaled, as well as delays for flip-flops, routing buffers and wire line delays.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2022", "sentence1": "On the other hand, any solution for U(i, S) when the current state is indeed i­-reachable is non-spurious.", "sentence2": "when a solution is found the reachability checking procedure of [9] is used to determine if the current state is i-reachable or not.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2023", "sentence1": "It is expected that some of the unreachable states may be intended to remain invalid by the specification.", "sentence2": "the design being debugged is expected to satisfy some invariant properties required by the specification.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2024", "sentence1": "This includes all solutions that make the target state reachable one cycle after a k-reachable state, and therefore the solution set of Algorithm 2 is a superset of the solution set of Algorithm 1.", "sentence2": "the solution set of Algorithm 2 can include solutions that would be discarded in Algorithm 1.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2025", "sentence1": "While unreachable code may be a symptom of an unreachable state(s) and vice versa, in practice they often manifest themselves separately.", "sentence2": "automated techniques to aid the engineer in fixing a design when a state is shown to be unreachable are of paramount importance to reduce the verification burden and improve the design cycle.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2026", "sentence1": "Whilst there has been a lot of interest in researching new architectures for GALS [1][2][3], there have been few attempts at providing synthesis solutions for GALS communication.", "sentence2": "generation of GALS from specifications has been limited to hardware description languages such as Verilog, VHDL [4] or synchronous programming languages such as C or ES-TEREL [5].", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2027", "sentence1": "To implement the model in circuit simulators, SPICE compatibility is essential.", "sentence2": "both Vaccess and RM show wide spreads.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2028", "sentence1": "Therefore, each individual bit-cell can have a smaller area with increased write failure; yet, extra bits for ECC are inserted to compensate for the increased failure rate.", "sentence2": "unlike the previous case, the individual bit-cell is positioned at the write and read dominated region, therefore, decreasing the barrier height would decrease the failure rate.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2029", "sentence1": "Note that for a fixed effective bit-cell area, the failure rate increases with the barrier height because JC is increased.", "sentence2": "increasing the effective bit-cell area increases JWR and reduces the write failure rate.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2030", "sentence1": "As seen in Fig. 1 (a,b), the STT-MRAM bit­cell consists of an MTJ and a MOSFET.", "sentence2": "the techniques published in literature are either focused on the optimization and failure mitigation at the bit-cell level [8,9,10], or on memory architecture design techniques to improve STT-MRAM yield [11,12].", "label": "neutral", "domain": "Hardware"}
{"id": "test_2031", "sentence1": "On the other hand, read-disturb failures occur because IRD is such that JC is exceeded during read operations and the stored data is accidentally overwritten.", "sentence2": "increasing the access transistor width to mitigate write failures also increases the read-disturb failures.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2032", "sentence1": "Write failures occur because JWR falls below JC.", "sentence2": "read-disturb failures occur because IRD is such that JC is exceeded during read operations and the stored data is accidentally overwritten.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2033", "sentence1": "This is simply because of the power-line switching overhead which becomes more notable when the system has to be turned off and on frequently.", "sentence2": "efficient power-gating can be applied to a unit which has the following characteristics: first, it should have a low utilization in order to have a good potential of being turned off, and second, its idleness must not be fragmented so that the power-line switching overhead remains negligible [11].", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2034", "sentence1": "Existing research efforts in formal verification of asynchronous circuits have been divided into three broad camps: hazard-freedom or conformance checking, equivalence checking, and property verification.", "sentence2": "many techniques try to verify hazard-freedom/conformance of a gate-level implementation of a leaf-level CSP cell, a unique requirement to asynchronous design [10], [11].", "label": "entailment", "domain": "Hardware"}
{"id": "test_2035", "sentence1": "The SEND and RECEIVE are asynchronous modules modeled by the 3VL operators R and S defined in Figure 1.", "sentence2": "in each iteration, based on the value received on E, the RECEIVE primitive may or may not receive from L, but it always sends a value on R.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2036", "sentence1": "These leaf cells can then be implemented with a variety of asynchronous pipeline templates.", "sentence2": "by adding constraints to the source Verilog to tell the verification tool to not consider these unreachable states, a common practice in synchronous LEC flows [21], the tool proved equivalence in 31 seconds.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2037", "sentence1": "We propose a method for logical equivalence check (LEC) of asynchronous circuits using commercial synchronous tools.", "sentence2": "we verify the equivalence of asynchronous circuits which are modeled at the CSP-level in SystemVerilog as well as circuits modeled at the micro-architectural level using conditional communication library primitives.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2038", "sentence1": "Since each iteration can be treated as encoding an edge of the tree to a number, the resultant Prüfer code preserves the connection properties of the original tree.", "sentence2": "in addition to the labels on a labeled tree, a circuit graph contains supplementary circuit information.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2039", "sentence1": "Assist Feature Insertion Sub-resolution assist features (SRAFs) are commonly employed in industry, not only to reduce EPE, but also to improve process window.", "sentence2": "assist features are essential for isolated features ", "label": "entailment", "domain": "Hardware"}
{"id": "test_2040", "sentence1": "After stage 1 and 2, either all the EPE violations are fixed, or EPE cannot be further improved.", "sentence2": "instead of simulating with dose variation, we can approximate the image by using Ith = 0.216 on the intensity values simulated at nominal condition, which saves us one simulation in each iteration.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2041", "sentence1": "As discussed in the example this DQBF has no equivalent QBF prefix.", "sentence2": "there is no need for explicit elimination of the internal auxiliary variables resulting from the Tsetin encoding.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2042", "sentence1": "Since the scales of circuit design and macro blocks are now very large, mapping macro blocks is difficult and has become an infeasible tasks for engineers to perform manually.", "sentence2": "it is valuable to develop a robust methodology that can utilize macro blocks automatically and efficiently.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2043", "sentence1": "Transistor folding is a task [2] normally done first, sometimes also taking transistor placement into account.", "sentence2": "they are still viable for a placement instance of a few dozen of transistors.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2044", "sentence1": "This problem gets even harder when sporadic tasks are involved.", "sentence2": "to demonstrate scheduling for FPPNs, we consider a practically relevant subclass of FPPNs where the use of sporadic tasks is restricted.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2045", "sentence1": "This suggestion is exploited in our model, Fixed Priority Process Network (FPPN).", "sentence2": "what if the job arrives exactly on the boundary?", "label": "neutral", "domain": "Hardware"}
{"id": "test_2046", "sentence1": "InputA has priority over FilterA and NormA, and hence it is joined to both of them.", "sentence2": "in the latter case the edge is redundant due to a path from InputA to NormA.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2047", "sentence1": "All stages of the receiver are fully differential.", "sentence2": "this integrated solution does not contain any stage driven by singleended signals as opposed to discrete component solutions, that need to minimize the transistor count and only have access to off-the-shelf components, which are often not available with differential inputs and outputs.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2048", "sentence1": "The logical simulation time of the kernel is referred to as global simulation time.", "sentence2": "the components can keep track of their time using a local simulation time, which is usually defined as an offset to the global simulation time.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2049", "sentence1": "It is essential for companies to invest in making software models for meeting their time to market.", "sentence2": "the fastest model is not a good model if it does not accurately match or predict the final design reality.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2050", "sentence1": "The internal memory states '0 'and '1' of a CRS cell are indistinguishable at low voltages because state ‘0’ as well as state ‘1’ show a high resistance. ", "sentence2": "no parasitic current sneak paths can arise.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2051", "sentence1": "These sensors are chosen for data fusion as they have been widely used in previous activity recognition attempts.", "sentence2": "we are interested in using different classification algorithms to perform a detailed analysis of the raw data collected from eight physical activities including idle, sitting, standing, walking, running, cycling, walking upstairs and downstairs.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2052", "sentence1": "In this study, spectral peak tracks (SPTs) are explored to represent the trace of the advertisement call.", "sentence2": "the time and frequency boundaries of each acoustic event are first defined by the AED method; spectral peaks are then extracted from the area of detected acoustic events.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2053", "sentence1": "We assume a hardware cache whose block size, associativity, and number of sets are B bytes, W , and S, respectively.", "sentence2": "the number of bits to address a set in the cache is log S.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2054", "sentence1": "It also uses a new congestion detection metrics for updating local congestion information and is used to set the learning rate to keep the values as updated as possible.", "sentence2": "when the switch is not congested the learning rate is set to a minimum value.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2055", "sentence1": "Moreover, the servers need to be kept in an air-conditioned environment, which are known to consume more power.", "sentence2": "we need a system to smartly monitor and manage the power consumed and minimize the loss and wastage [8].", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2056", "sentence1": "This ever-growing environmental problem engages health risks and major complaints of annoyance on behalf of millions of citizens.", "sentence2": "sustainable urban planning needs to seriously take into consideration the task of mitigating environmental noise.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2057", "sentence1": "We have adopted a time window of 15 seconds (i.e. 40 se­quential segments for each recording) where the signal ex­hibits homogeneous behavior with respect to audio type (context and soundscape quality).", "sentence2": "within these time windows, both context and the soundscape quality is supposed to be stable homogeneous.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2058", "sentence1": "For the particular case of the soundscape quality estimation task, we have adopted an alternative approach (called indirect soundscape quality -ISQ): instead of direclty mapping points of the feature space to scalar estimates (typical regression), we have also implemented a meta-regression scheme. ", "sentence2": "we use the estimated levels for the three context classes as feature values in a meta-regression approach (see Figure 8).", "label": "entailment", "domain": "Hardware"}
{"id": "test_2059", "sentence1": "In order to measure the overall latency of a system with a player with a physical disability, a system was implemented using the Xbox 360 Kinect and a Windows computer running the Kinect SDK version 1.8.", "sentence2": "these are outliers, and the average of the absolute value of the deviation between drum and guitar was under 150ms, with times as low as 30ms.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2060", "sentence1": "The idea of projection-based denoising is not new and it has been effectively applied in image denoising, see e.g., [7] and [3].", "sentence2": "to the best of our knowledge it has not been utilized in denoising PSG data streams which are extremely heterogeneous.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2061", "sentence1": "Through our work, we show that UrJar has the potential to channel e-waste towards the alleviation of energy poverty, thus simultaneously providing a sustainable solution for both problems. ", "sentence2": "we present one such attempt, a backup power device - called UrJar - that seeks to simultaneously address the problems of proliferation of laptop battery e-waste, and the prevalence of energy poverty in developing countries.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2062", "sentence1": "In this paper we claim that understanding QoE in mobile devices is paramount for cellular network operators, and present the results obtained from subjective lab tests performed for popular end-user services accessed through smartphones. ", "sentence2": "we consider the following five well-known applications in mobile devices: YouTube dynamic and non-dynamic video streaming, Facebook, Google Maps (Gmaps from now on), Web browsing through Google Chrome, and WhatsApp.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2063", "sentence1": " FaFNoC achieves slightly better results than FON or FTDR-H for throughput and average hopcount, whereas FTDR achieves significantly better results for those values.", "sentence2": "the existing architectures FON and FTDR-H lose a huge amount of injected flits. ", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2064", "sentence1": " The inherent redundancy of NoCs can be used to tolerate such failures. ", "sentence2": "at FTDR and FaFNoC almost all injected flits arrive at their destinations.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2065", "sentence1": "This makes de.ection routing attractive for small, energy e.cient and fault tolerant NoCs.", "sentence2": "only few fault-tolerant de.ection-routing based router architectures exist.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2066", "sentence1": "Because the mesh, Cmesh, and butter.y NoCs provide diverse paths between L1s and individual L2s, they lead to higher utilization of the CU.", "sentence2": "the Crossbar, Clos and C-Crossbar, which at some point combine the tra.c to/from the L2s, achieve lower performance.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2067", "sentence1": "the zero-load broadcast latency of BoWNoC is scalable as it remains constant when varying the number of cores.", "sentence2": "the MAC protocol must also be scalable (which is a challenge in itself ) in order to retain such latency advantage.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2068", "sentence1": "In a WNoC, a transmitting antenna radiates RF signals at a given frequency, which propagate throughout the chip and may be received by any other a­tenna tuned to the same frequency.", "sentence2": "all antennas tuned at the same frequency channel share the medium similarly to in a bus.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2069", "sentence1": "Being located in the far field of the antenna, the core environment does not necessary have an impact on the gain and matching of the antenna.", "sentence2": "the environment can have an effect on the link between the transmitting and receiving antennas by causing increased path and polarization losses, and limiting the data rates.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2070", "sentence1": "Different kinds of attacks like node replacement, node movement or man-in-the-middle were considered for investigation.", "sentence2": "he suggested that RSSI can be used in devices which focus on energy consumption [2].", "label": "neutral", "domain": "Hardware"}
{"id": "test_2071", "sentence1": "In execution activity are instrumented the desired adaptation acts by actuators or effectors.", "sentence2": "the highly changeable environment of WSN has Pi-ADL as a suitable candidate for representing the RA.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2072", "sentence1": "To evaluate the RA, ProSA-RA suggests using a checklist-based inspection approach named FERA (Framework for Evaluation of Reference Architectures) [30].", "sentence2": "we followed the recommendation and used FERA to conduct an evaluation in order to verify: (i) completeness of general information related to the construction and content of the architectural views; (ii) adequacy for releasing of the architectural description of our RA; and (iii) viability and change possibilities of our RA.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2073", "sentence1": "In [9], we found different approaches to provide autonomic behavior to WSNs, but they are just in preliminary stages and have still several open issues.", "sentence2": "we noticed a lack of well-defined reference architecture designs to support the autonomy of WSN.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2074", "sentence1": "Empirically, we find that the 10th percentile of ranges estimates gives the least average error.", "sentence2": "the 10th percentile aggregate of the 27 range estimates is used as the true range at each anchor.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2075", "sentence1": "Moreover they all produce an error of, at most, one bit.", "sentence2": "when one of the above events occurs, a bit error may or may not be produced.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2076", "sentence1": "The need for providing com­munication strategies in these challenging environments where electromagnetic waves cannot be used, has brought attention of researchers towards molecular communications.", "sentence2": "the possibility of moving very small amounts of matter inside microflu­idic systems can be exploited to exchange information between micro-and nano-scale devices.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2077", "sentence1": "Inside this droplet, the luminescent process takes place or not, according to the presence or absence of signal- Figure 3: Graphic representation of the adopted encoding approach ing molecules.", "sentence2": "downstream the signaling molecule detector, a sequence of luminescent and non luminescent droplets is produced, where the luminescent droplet represents the bit 1 and the non-luminescent droplet represents the bit 0 .", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2078", "sentence1": "For instance, enabling multiple (more than two) Graphical Processing Units (GPUs) in a single system is nowadays problematic as the interface between them becomes a bottleneck.", "sentence2": "providing efcient communications between various intra-computer components, such as Central Processing Units (CPUs) and GPUs, is a hot research topic nowadays.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2079", "sentence1": "Following the Beer-Lambert law the absorp­tion component of the path loss is modeled as exponential function e-K(f)d, where K(f) is absorption coefficient, d is the separation distance.", "sentence2": "well-polished metal surfaces or metal coating mirrors could act as THz reflectors [9, 10].", "label": "neutral", "domain": "Hardware"}
{"id": "test_2080", "sentence1": "As one could observe, the water has exceptionally strong absorption in the THz band.", "sentence2": "ammonia is characterized by rare absorp­tion lines at isolated frequencies.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2081", "sentence1": "The use of technical refinement is playing an important role in the process of optimizing training efficiency and improving results of athletes.", "sentence2": "we aim to develop an online monitoring and feedback tool to follow the swimming training using a single waterproofed and wireless motion sensor attached on the swimmers center of body (back).", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2082", "sentence1": "According to the random channel selection scheme, the channels are arranged in a random sensing order and the SUs transmit in the first vacant channel they sense.", "sentence2": "[7] proposes an intuitive channel sensing order arranging the selection of the channels in a descending order of their a priori known availability probabilities (ascending order of their occupancy probabilities).", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2083", "sentence1": "A set of experimental measurements is presented aiming to demonstrate the experimental framework and verify the LTE/LTE-A standards.", "sentence2": "three experiments are demonstrated according to the Reference Measurement Channels (RMCs) defined by 3GPP [5], [6].", "label": "entailment", "domain": "Hardware"}
{"id": "test_2084", "sentence1": "In this paper, an experimental framework for studying and measuring LTE and LTE-A is presented, consisting of high-end laboratory equipment and software tools.", "sentence2": "an arbitrary signal generator is exploited i.e., the Le Croy 1102D together with the ArbStudio software, providing two channels necessary for MIMO.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2085", "sentence1": "The C-to-ADA translator is based on the front end of a C compiler. ", "sentence2": "it performs all early stages of a compiler, including lexical, syntactical and semantical analyses.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2086", "sentence1": "The C programs for this benchmark are listed in Table 1.", "sentence2": "heuristics are applied on the HLS transformations", "label": "neutral", "domain": "Hardware"}
{"id": "test_2087", "sentence1": "In Proceedings of the 9th international conference on Ubiquitous computing (UbiComp '07) (pp.", "sentence2": "note that for the same container, the impulse response differs for the different content levels.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2088", "sentence1": "Surprisingly, we have also noticed subtitle files with differing pace.", "sentence2": "even when they are properly aligned at the first caption entry, they eventually get out of sync with each other.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2089", "sentence1": "For secondary analysis, however, the procedure is generally fixed for all samples that come off the sequencers.", "sentence2": "further downstream analysis involves computational algorithms that directly answer the scientific questions at hand and are varied in their computational requirements.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2090", "sentence1": "The proposed hardwaresoftware technique leads to a mean improvement of 18% in application performance and a mean reduction of 26% in misses over a shared LLC managed by the Least Recently Used replacement policy for a set of input-annotated taskparallel programs using the OmpSs programming model implemented on the NANOS++ runtime.", "sentence2": "the state-of-the-art thread-based partitioning scheme suffers an average performance loss of 2% and an average increase of 15% in misses over the baseline.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2091", "sentence1": "The insight behind FECC follows recent works which observe that compression at the cache-block granularity can free up enough space for other information [17, 18]; we use this free space for storing the ECC codes. ", "sentence2": "when compression succeeds, FECC can match the performance of a conventional ECC organization that uses dedicated ECC memory devices with less, or even zero, dedicated redundancy.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2092", "sentence1": "Sim­ilar to the use of difference coding for integer values, we can compress the sign and exponent with simple subtraction assuming that their val­ues exhibit locality.", "sentence2": "other approaches exist, similar to VECC, that em­bed ECC data elsewhere in the memory space to allow a more flexible use of non-ECC memory modules [6, 8].", "label": "neutral", "domain": "Hardware"}
{"id": "test_2093", "sentence1": "Clearly, the uneven distribution of SBEs among GPUs make this analysis challenging.", "sentence2": "they are an indicator of error resilience of the hardware and have implications toward the ECC page retirement error as described earlier.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2094", "sentence1": "Traditionally, graphics processing applications, such as video games and image processing applications, have utilized GPUs for faster processing and rendering.", "sentence2": "with the increased thrust towards general-purpose computing on graphics processing units (GPGPU), GPUs have started to influence high performance computing (HPC) systems as well.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2095", "sentence1": " In this section, we want to understand the relationship between single bit error (SBE) and GPU resource utilization.", "sentence2": "we investigate the correlation between GPU core hours, number of nodes and memory consumption with SBEs.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2096", "sentence1": "In addition to reporting the single bit er­rors, nvidia-smi output also includes double bit and ECC page retirement related errors.", "sentence2": "we use both the­se methods of data collection to quantify and analyze the characteristics of GPU errors.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2097", "sentence1": "Our double bit error analysis so far is based on the console logs.", "sentence2": "titan, the world’s second fastest supercomputer, consists of 18,688 NVIDIA Tesla K20X GPUs that computational scientists use routinely to perform scientific simulations and data ana­lysis.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2098", "sentence1": "A huge number of VR researchers, programmers, and engineers have been trying to tackle this issue from many aspects.", "sentence2": "reproducing the human visual sense in a remote place has been gathering attention recently, backed by many releases of sophisticated devices, omnidirectional cameras, and head mount displays (HMDs).", "label": "entailment", "domain": "Hardware"}
{"id": "test_2099", "sentence1": "At the middle of the long stationary period around 280-440 sec­onds, there was another train having a brief stop at the opposite side of the platform.", "sentence2": "stop probability for the stationary train temporarily dropped due to the magnetic noise, caused by acceleration/deceleration of the neighboring train.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2100", "sentence1": "All data was recorded using the same smart­phone, but in independent sensing runs over the course of one year with long pauses between them.", "sentence2": "data was sub­ject to possible systematic errors due to varying temperature, slightly different experimental setup and even changed methods of data collection: The first few datasets were recorded with an older version of the sensing app, which heavily used automatic white balance and exposure compensation.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2101", "sentence1": "In this section, we survey these systems, noting the effect that these choices have on overall size and power draw.", "sentence2": "we survey three commercial meters Kill-A-Watt [6], Watts Up [17], and Belkin Conserve Insight [3] and three research meters ACme [26] (two different versions), Monjolo [23], and Gemini [20] alongside PowerBlade (this work).", "label": "entailment", "domain": "Hardware"}
{"id": "test_2102", "sentence1": "Hall effect sensors work by measuring the deflection of electrons in a conductor exposed to a magnetic field (like the one generated by a current).", "sentence2": "powerBlade optimizes for size by using a power supply design that does not require an IC [21].", "label": "neutral", "domain": "Hardware"}
{"id": "test_2103", "sentence1": "In this section, we discuss some limitations of the current design, explore some possible workarounds, and propose some future directions for improvement.", "sentence2": "the accuracy could be further improved, the wireless system could be better utilized, and interval data could be collected with periodic timestamps.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2104", "sentence1": "Even though smartphones are typically sensitive to ultrasonic sound, their speakers are highly directional in those frequencies, which lead [20] to use audible frequencies.", "sentence2": "in situations where the user blocks one or more transmitters while walking, or when a NLOS signal is detected, the system cannot update the location estimate.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2105", "sentence1": "In concordance to the 60 frame long transient phase, the CCC has a maximum at also approximately 60 frames.", "sentence2": "the descriptor is from now on referred to as PHOG-TOP.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2106", "sentence1": "One important adaptation is using Lin s concordance correlation coefficient (CCC) [17] rather than Pearson s correlation coefficient to measure performance.", "sentence2": "3.1 Audio Features The extraction of linear predictive coding coefficients (LPC) is an auto regressive approach, where the n th sample of a time series is approximated using a function of the p preceding samples [1]: LPC are still widely applied in speech processing, for example in speech recognition and speech synthesis.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2107", "sentence1": "Several mixnet and onion routing systems have sought to make traffic analysis more difficult using cover traf­fic, i.e., fake traffic on each communication link [4, 20], or delaying messages [24].", "sentence2": "it has been shown that all these schemes still reveal information after multiple rounds of observation [26].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2108", "sentence1": "To address this attack, Vuvuzela uses a mixnet approach.", "sentence2": "all requests are recursively encrypted under the public key of each server in Vuvuzela’s chain.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2109", "sentence1": "Although cover traffic hides the exact number of dead drops accessed once or twice, an adversary can still tell roughly how many people are talking, by looking at the number of dead drops accessed twice and subtracting the average amount of noise.", "sentence2": "in the above configuration, clients use an average of 12 KB/sec (adding up to 30 GB over a month of continuous use, which may be high for a mobile phone with metered data service).", "label": "neutral", "domain": "Hardware"}
{"id": "test_2110", "sentence1": "Vuvuzela aims to provide point-to-point messaging between users in a way that is private in the face of a strong adversary, who can observe and tamper with the entire network and all but one of Vuvuzela's servers.", "sentence2": "an adversary should not be able to distinguish between a scenario where two particular users are communicating, and a scenario where they are not, even after interfering with the system.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2111", "sentence1": "The multiplicative ε provides plausible deniability: any event observed if, say, two specific users are talking has within eε times the probability if they were doing something else (e.g., within 1.1× for ε = 0.1 and within 2× for ε = ln 2), so users always have a plausible \"cover story\".", "sentence2": "δ covers events that might have zero probability under some actions but happen under others.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2112", "sentence1": "For instance, scaling up to 2 million users increases the latency from 37 to 55 seconds.", "sentence2": "in our analysis we assumed 10minute dialing rounds, which means a client must wait on the order of 10 minutes to start a conversation.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2113", "sentence1": "If the adversary controls the third server, he can now figure out whether Alice and Bob are talking!", "sentence2": "if Alice and Bob are communicating, the third server will see two exchanges for the same dead drop; otherwise, it will not.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2114", "sentence1": "Vuvuzela aims to provide point-to-point messaging between users in a way that is private in the face of a strong adversary, who can observe and tamper with the entire network and all but one of Vuvuzela's servers.", "sentence2": "finally, to not let noise affect the clarity of the graphs, we configure servers to always add exactly µ noise, rather than sampling the Laplace distribution; this produces the same average results with less variance.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2115", "sentence1": "Ekho: Realistic and repeatable experimentation for tiny energy-harvesting sensors.", "sentence2": "in this paper we describe Enspect, an open-source hardware/software tool which simpli.es the design of energy harvesting sensing systems by assisting in the specification of harvesting and storage devices.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2116", "sentence1": "For AF without SD link For AF without SD link when optimum power consumption is averaged over fading channel coofficients, the minimum average power consumtion takes places when relay node lies midway between source and destination terminal as plotted in Fig. 2.", "sentence2": "with x axis denoting position of relay terminal and taking   power plots are shown for all the four cases.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2117", "sentence1": "In a practical scenario, signal transmission through the wireless channel suffers from fading, which results in unreliable reception of the signal at the receiver.", "sentence2": "it is either positive or negative in R, then the autocorrelation function is nonnegative.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2118", "sentence1": "It is observed that if we increase the number of words present in a corpus, the processing time increases and this increment is larger in bigger corpus size as compared to smaller size corpuses.", "sentence2": "keeping smaller corpus size effectively reduces the overall computation time during clustering.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2119", "sentence1": "Since transmitter baud rate is 18450 bps and receiver baud rate is 20000 bps , the percentage difference in their baud rates is 7.750% which is more than 4.575%.", "sentence2": "framing error is generated at the receiver.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2120", "sentence1": "First, the processor chip must be redesigned to deal with having a multitude of TSVs driven through it (which has implications in terms of engineering effort, tools support, physical design, etc.).", "sentence2": "a relatively conventional chip design can be used for 2.5D stacking as the processor die has no TSVs and does not need to undergo die thinning (risking cracking and other mechanically-related yield issues).", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2121", "sentence1": "Wide I/O [20] has low speed and thus very large bus width with only one-to-two channels sup­ported.", "sentence2": "wide I/O 2 [24] has increased the number of channels to four-to-eight allowing the bus width to re­duce (with larger prefetch) and increasing the frequency.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2122", "sentence1": "The previous analysis shows that increasing the number of channels is extremely beneficial for performance reasons.", "sentence2": "a problem with increasing the number of channels is the reduction in DRAM density.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2123", "sentence1": "However, the number of cycles in which the bus is actually transferring data (the data cycles) for a given transaction size remain constant.", "sentence2": "in the case of consecutive page misses, in which the next transaction can be sent after a minimum of tras cycles, only a few of the bus cycles will be trans­ferring data.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2124", "sentence1": "On average, the combined power consumption of the processor and DRAM keeps at the same level", "sentence2": "for the adoption of MVX, several implementation issues have to be considered, which are dis­cussed in this section.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2125", "sentence1": "For high-MLP applications, the bandwidth upgrade will reduce contention among concurrent mem­ory requests, reduce memory latency, and improve performance.", "sentence2": "3D-stacked DRAMs cannot reduce idle-system memory latency.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2126", "sentence1": "TSV links shorten interconnection paths and reduce connectivity impedance.", "sentence2": "data can be moved at higher rates with lower energy-per-bit.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2127", "sentence1": "The resulting extremely high access parallelism (at the expense of loss of unneeded flexibility) offers enormous power-performance-cost benefits when relevant.", "sentence2": "the data flow graph and the assembly plant analogy expose a very important fact: any given data item is generated by one entity and is usually only required by very few others.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2128", "sentence1": "An illustrative example is a program operating on a stream of input data, and a situation wherein the result computed by a given ALU (instruction in the program) is used as an operand by several others.", "sentence2": "the arrival times of the remaining operands required by each of the consuming ALUs differ and are moreover not known exactly in advance, as they are data dependent.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2129", "sentence1": "The similarity between spatial computing and an assembly line and with data routing in general, calls for examining the relevance of various buffering, queuing and routing schemes and constructs used there to the organization and interconnection of operand memory in a spatial computer.", "sentence2": "it is important to create a set of such constructs and to implement them efficiently as building blocks in the hardware (just like lookup tables, flip flops, multipliers and connections).", "label": "entailment", "domain": "Hardware"}
{"id": "test_2130", "sentence1": "When one chip fails or there are multiple errors in a chip, Chipkill-Correct and E-ECC Scheme 1 can correct these errors without additional memory reads.", "sentence2": "v-ECC, LOT-ECC and Multi-ECC all require additional reads to correct errors.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2131", "sentence1": "In case of row failure due to random errors, Scheme 2 detects the errors with 99.9986% probability (Section 5.5) and relies on higher level protection mechanisms such as roll back to recover the data.", "sentence2": "in such cases, Scheme 2 will incur performance and energy loss due to use of the higher level protection.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2132", "sentence1": "In addition, to evaluate the multi-core system performance, we create two 4-core multiprogrammed workload mixes of the aforementioned SPEC2006 benchmarks to realistically model the multiprogrammed application execution scenarios.", "sentence2": "next, we present the various existing ECC schemes (in Section 2.3).", "label": "neutral", "domain": "Hardware"}
{"id": "test_2133", "sentence1": "This code can correct an erasure from a chip that has been marked faulty and can detect one more random error in another chip.", "sentence2": "for higher reliability, it is important that the random error in the second chip be corrected and so a stronger code is necessary.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2134", "sentence1": "If so, it activates single error correction; otherwise, it declares there are errors.", "sentence2": "in this section, we analyze the reliability of the compet­ing schemes.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2135", "sentence1": "In this paper, we aim to reduce the cell-spacing below the WD-free distance of ∆, and apply innovative ideas to co­design the layout and coding to eliminate WD.", "sentence2": "it is important to realize that modifying the cell layout is con­strained by the physical limits imposed by the underlying technology node.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2136", "sentence1": "Hence, a higher number of dies per wafer for a given density directly translates into lower cost per die.", "sentence2": "the final cost of a die depends not only on die count, but also on the die yield since die cost is directly proportional to the inverse of yield [9].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2137", "sentence1": "However, this approach is increasingly at odds with technology scaling and the effect of PV due to a wider range of latency behav­ior, particularly in a cell's access transistor and capacitor.", "sentence2": "the impact from PV grows stronger with decreasing transistor and capacitor size, required for increasing chip density, which may cause more yield loss due to operational timing violations.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2138", "sentence1": "We also analyzed the trade-off between relaxed timing and application performance.", "sentence2": "how much does the timing have to be relaxed to regain enough yield and what is the performance cost?", "label": "entailment", "domain": "Hardware"}
{"id": "test_2139", "sentence1": "Rather than relying on conventional strategies, better-than worst­case design approaches, which relax timing and manage performance, will be required, spanning DRAM design, architecture (mem­ory controller) and even the system software.", "sentence2": "we advocate in this paper for a new soft yield strategy in which memory timing constraints are allowed to vary by location to rescue chips that would have failed under strict global timing constraints.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2140", "sentence1": "Hence, the larger the block size, the more likely for blocks to be downgraded to slow blocks.", "sentence2": "it may be possible to architect blocks to define new logical blocks composed of fast, normal, and slow physical blocks.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2141", "sentence1": "In particular, several memory sub-system parameters can be tweaked, possibly to miti­gate the performance loss from relaxed timing.", "sentence2": "the memory system could potentially be provisioned differently to enable relaxation of the timing; with small degrees of PV, it may be possible to adopt relaxed global timings with appropriate configuration of memory sub-system parameters, such as redundancy, buffering and bank parallelism.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2142", "sentence1": "GPUs appear to be ideal accelerators for streaming computations: with their many processing cores, today’s GPUs have 10X the compute power of modern CPUs, and they have close to 6X the memory bandwidth of modern CPUs,1 yet are priced as commodity components.", "sentence2": "a number of issues had until recently prevented effective acceleration in practice.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2143", "sentence1": "These efforts have shifted the primary bottleneck preventing higher GPU core utilization from the PCIe link to the GPU­side memory hierarchy.", "sentence2": "three factors currently prevent further improvements in core utilization.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2144", "sentence1": "The decision to use thread-private cache segments is based on the fact that inter-thread data sharing is rare in the streaming applications we are target­ing.", "sentence2": "the threads mostly process data independently in disjoint locations of memory.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2145", "sentence1": "For accesses to non-cached data structures, the performance overhead entails the exe­cution of four extra machine instructions.", "sentence2": "accesses to cached data structures incur signifcantly more overhead in some cases; e.g., when evicting a cache line.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2146", "sentence1": "We decided on using 16-byte cache lines after experimenting with different cache line sizes — see Section 4.5. ", "sentence2": "based on our experiments, the monitoring phase accounts for less than 1% of this overhead.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2147", "sentence1": "A good design needs an in-depth understanding of end-user workloads.", "sentence2": "designers rarely get insights into end-user work­loads because of the proprietary nature of source code or data.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2148", "sentence1": "Workload cloning is an emerging approach that can bridge this gap by creating a proxy for the proprietary work­load (clone).", "sentence2": "the DRAM subsystem is organized in a hierarchical fashion consisting of channels, ranks, banks, rows and columns as shown in Figure 1.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2149", "sentence1": "We argue that the overhead of reuse distance in main memory is negligible and feasible for online estimation for the following two reasons: 1) The tra.c in main memory is 40-100× smaller than in the cache system.", "sentence2": "the estimation overhead can be drastically reduced.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2150", "sentence1": "Compared to PCache, HRank causes 7.1x more data accesses in the PCM.", "sentence2": "the energy consumption of HRank is more than the PCache in the pF3D case.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2151", "sentence1": "A perfectly unequal distribution (i.e. only one memory page is ever accessed) would have a Gini coefficient of 1.0.", "sentence2": "additionally, mem­ory vendors see new architectures as a way to differentiate themselves and offer higher-margin products.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2152", "sentence1": "35% of pages account for almost 75% of memory requests.", "sentence2": "minife, a .nite element application, has a more equal distribution (the top 35% of pages only account for 45% of accesses).", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2153", "sentence1": "Broadly speaking, each of the 10 workloads presented in Section 5 has high memory traffic.", "sentence2": "but hardware caches are not without their challenges.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2154", "sentence1": "Traditional OS page cache management policies operate as demand paging replace­ment policies.", "sentence2": "when a memory access occurs to a page that is not resident in cache, the replacement policy decides which victim page to evict from the cache to make room for the freshly-­accessed page.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2155", "sentence1": "We use another metric from Weinberg for computing spatial locality based on the average length of non-zero strides across an application within a lookback window W=32 that captures address strides. ", "sentence2": "within the previous W addresses, find the one with the minimum absolute distance to the current one and call it the stride length.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2156", "sentence1": "When the PCM has more levels, herniated hash tables can yield more benefits by deeper prefetching and more overlap between computation and memory accesses.", "sentence2": "since latency is exponential with respect to the depth of the level accessed, there is a significant overhead when the highest level is accessed from PCM.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2157", "sentence1": "In each case, the data shows two columns, one for only accelerating the Txlog and one for accelerating both TxLog and BinLog.", "sentence2": "direct mmap directly accesses data in PM without caching in DRAM [7].", "label": "neutral", "domain": "Hardware"}
{"id": "test_2158", "sentence1": "Even when remote, the PM configuration still delivers better performance than local flash when used over Infiniband.", "sentence2": "as latency increases substantially with Ethernet, the application performance of remote PM drops dramatically to become far less than that of local flash.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2159", "sentence1": "Identification and definition of guidelines for manual composition of convergent services are problems that have been investigated for years [18]; more recently, automation of different phases of convergent composition has been actively researched.", "sentence2": "we have developed a framework called AUTO [3][4][5], which aims at supporting the automated composition of convergent services using automated planning for service composition and natural language processing for user request processing.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2160", "sentence1": "The data in Figure 9(a) assumes an ideal MPPT implementation.", "sentence2": "models with better accuracy can be obtained with fewer measurements from a relatively uncontrolled en­vironment.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2161", "sentence1": "For powering energy-neutral WSN, PV energy harvesters generally offer the highest volumetric power output [2], and power conversion and maximum power point tracking (MPPT) techniques for solar energy harvesters are relatively well understood.", "sentence2": "the extent to which available energy varies over time can cause problems for system designers. ", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2162", "sentence1": "Systems-of-Systems (SoS) result from associating independent, complex systems for fulfilling given missions.", "sentence2": "soS require software architectures that can cope with their dynamic, critical nature.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2163", "sentence1": "A contract expresses behavioral properties as well as specific policies and constraints that are necessary for coordinating constituents in an SoS.", "sentence2": "furthermore, we identified desirable features for tools associated to ADLs for SoS.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2164", "sentence1": "Second, the mediator building block describes an architec­tural element that manages the interaction among heteroge­neous constituents within an SoS and promotes interoperabil­ity.", "sentence2": "mediators not only transmit data (as connectors in traditional systems), but they are also able to process such exchanged data and to coordinate the interactions among con­stituents.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2165", "sentence1": "As UML is a well established standard for modeling soft­ware systems, several tools currently support it.", "sentence2": "the Papyrus Modeling Tool is an open-source graphical editing tool available as an extension for Eclipse.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2166", "sentence1": "CML is a formal language designed for supporting SoS engineering.", "sentence2": "cML models could be derived from SysML block de.nition and state machine diagrams, thus enabling to describe the combined semantics of constituents in an SoS as synchronous events.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2167", "sentence1": "Some examples of formal and semi-formal ADLs identified in these studies are UML, SysML, CML, and X-UNITY.", "sentence2": "the combination of formal and semi-formal ADLs has also been proposed for describing SoS.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2168", "sentence1": "X-UNITY is a formal language that addresses the descrip­tion of interfaces of constituents and SoS.", "sentence2": "constituents in a X-UNITY description present internal variables (called context) and exposed variables that are visible to other constituents.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2169", "sentence1": "WZeroRPC is an open source platform that takes a different approach by offering a workflow-based execution model that can be embedded within conventional programming paradigms.", "sentence2": "it provides support for executing a directed acyclic graph (DAG) workflow, executed across a distribution of workers, along with the dependencies between them, whilst providing a Python interface for coordinating execution and data dependencies between multiple workflow instances.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2170", "sentence1": "In Section 4 we shall present WZeroRPC, which addresses this issue, and also explain its relationship to other WMS more generally.", "sentence2": "we shall first explain the context within which this work has been carried out, namely the SWITCH project.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2171", "sentence1": "Although cloud environments are capable of providing virtualized, elastic, controllable and high quality on-demand services, they still lack the ability to control the selection and configuration of infrastructural components in response to changing requirements and environmental pressures.", "sentence2": "current Cloud environments still do not have the tools and application programming interfaces that would allow the developers to exert such control on the underlying infrastructure in an intelligent, semi-autonomous manner.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2172", "sentence1": "Each node has an ID that must be unique on this server.", "sentence2": "when requesting the telemetry server using a node ID, this node (and its subnodes depending on the request) is used for the response or none if this ID is unknown.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2173", "sentence1": "Despite the modularity of existing workflow engines, described in Section 2, workflow systems generally provide a language to model the entire process and logic defining the interaction of components.", "sentence2": "dynamic interactions happen within the workflow language and structure of the workflow systems, which means that in order to create a dynamic data workflow, a developer would: Need to learn to program using the workflow language of a system. ", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2174", "sentence1": "Each Channel represents a semantic search and therefore is processed independently of other channels.", "sentence2": "multiple Channels can exist concurrently and also channels can grow rapidly as the interest in a particular topic increases by the users of the network.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2175", "sentence1": "Though designed specifically for minimising transmission times, SPDY also compresses and in some cases eliminates request and response HTTP headers.", "sentence2": "given that headers range in size from 200 bytes to 2 KB, its ability to reduce overall data traffic is limited [7].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2176", "sentence1": "This permits clients to retrieve precisely the required data by specifying filters on data items.", "sentence2": "we evaluate EDGEREDUCE using three real-world mobile client applications for Twitter, Groupon and Yahoo!", "label": "neutral", "domain": "Hardware"}
{"id": "test_2177", "sentence1": "The nRF24L01+ transceiver has several configurable parameters that can be tuned for the specific applications.", "sentence2": "its most configurable parameters that impact the energy efficency of the communication are: transmission speed, transmission power, payload size, activation of acknowledgments.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2178", "sentence1": "Power gating without retention FFs will dump its state into memory, which helps eliminate logic SER during gating.", "sentence2": "power gating with retention FFs will increase SER as FFs in retention mode are more vulnerable to soft errors.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2179", "sentence1": "The hardware evaluation platform has to be general enough for targeting different hardware configurations as well as fast enough for running system software.", "sentence2": "implementing actual hardware for each of the configuration is impractical and simulation-based approaches are not feasible.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2180", "sentence1": "In order to use MVX instructions, we require no changes to source codes.", "sentence2": "the code needs to be recompiled in order to make use of the MVX instructions.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2181", "sentence1": "Its evolutions as DDR 2, DDR 3, DDR 4 and so forth generally increased the I/O frequency by increasing the data burst capability and bus operating frequency.", "sentence2": "the organization of a DDR DRAM device in all versions suffered few architectural mod­ifications.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2182", "sentence1": "The branch bitmap maintains the knowledge of basic block boundaries within a block, al­lowing for providing the instruction fetch unit (L1-I), with multiple instructions to fetch in a single lookup.", "sentence2": "this is a rare event when an instruction prefetcher with high miss coverage and timeliness is employed.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2183", "sentence1": "Hence, we proposed to design accelerators that instanti­ate the APIs of these commonly used libraries in hard­ware.", "sentence2": "a hardware-software co-design ap­proach is used to target memory-bounded operations within commonly used high performance libraries (e.g., BLAS and FFTW) for portable energy efficiency.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2184", "sentence1": "The overview of our solution is shown in Figure 6.", "sentence2": "memory-bounded opera­tions (e.g., scalar product, general matrix vector multiply, and fast Fourier transform) bene.t to a lesser de­gree on accelerators.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2185", "sentence1": "Directive-based programming models such as OpenACC [50] are proposed to ease the programming burden of specialized accelerators.", "sentence2": "the end result is Memory Accelerated Library (MEALib).", "label": "neutral", "domain": "Hardware"}
{"id": "test_2186", "sentence1": "Since all elements of a given data type share the same expected range of values (as discussed in Sec. 4.1), elements with relatively smaller values (e.g., interest rates) become overly susceptible to approximate similarity, leading to some inaccuracy. ", "sentence2": "all of these overheads are faithfully modeled in the energy and area results in Sec. 5.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2187", "sentence1": "Pages that have only been accessed twice will have only one delta available and will be restricted to using the first DPT table for prediction.", "sentence2": "as the page is repeatedly accessed, and more history is accumulated, DPT lookups may produce matches in more than one of the DPT tables.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2188", "sentence1": "For the LPDDR3based point-to-point system, since all of the data are supplied by a single chip, it is easy to apply a data layout similar to what is shown here.", "sentence2": "this paper presents MiL (More is Less), a novel data communication framework built on top of DDR4, which exploits the data bus under-utilization caused by DRAM timing constraints to selectively apply sparse codes, thereby reducing the IO energy without compromising system performance.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2189", "sentence1": "However, current hypervisors already track the sharing status of each page for memory management.", "sentence2": "for vCache, its hypervisor just stores the already tracked status bit (1 bit) in the page table entry.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2190", "sentence1": "Due to the workonly/allocation policy with vCache (geo-mean of three conserving policy of vCache, VM1 can exploit unallocated/unused capacity along with its allocated vLLC and thus vLLC shows higher performance than pLLC while preserving VM2 s performance.", "sentence2": "vM1 can exploit more capacity before VM2 fills its al located capacity, since VM2 does not fully utilize its allocated capacity.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2191", "sentence1": "In terms of normalized LLC misses, vCache also preserves the effectiveness of the page coloring schemes for all workloads.", "sentence2": "such an illusion of resource dedication has not been supported for the last-level cache (LLC), although the LLC is the largest on-chip shared resource with a significant performance impact.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2192", "sentence1": "For virtualized systems, the hypervisor may directly enforce page coloring for VMs or applications to control cache allocation, but enabling vCache does not allow such hypervisor-based page coloring. ", "sentence2": "the rationale for vCache approach is to allow independent cache managements by the guest OS and hypervisor.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2193", "sentence1": "Consequently, sLLC shows higher performance improvement than pLLC for VM1. ", "sentence2": "for such control, each cache line keeps a VM identifier to identify the VM that loaded the cache line.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2194", "sentence1": "One simple method to provide a VM with an isolated LLC is to dedicate fixed physical cores and LLC slices to a single VM; a physical core is associated with a slice of LLC (e.g., a core has a 2MB LLC slice in an octa­core processor with 16MB LLC [30]).", "sentence2": "this approach is neither effective nor flexible.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2195", "sentence1": "A key role of machine virtualization is to provide an illusion that each workload runs on a dedicated machine, while sharing machine resources with other co-executed workloads [1].", "sentence2": "such an illusion should be rigorously provided for utility computing that is typically offered by the public clouds where each user desires his/her workload to fairly run on the paid resources.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2196", "sentence1": "Such a page coloring technique can allow the OS to evenly spread accesses across LLC sets or reduce interference amongmultiple applications sharingthe LLC byassigninga suitable LLC region for each application.", "sentence2": "it is widely supported by the mainstream operating systems such as Solaris, FreeBSD, netBSD, and Windows NT [12, 19, 13, 20, 17].", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2197", "sentence1": "V-SR provides isolation be­tween physical cores and the coherence scheme by map­ping logical sharers used in the coherence scheme into physical cores in the system.", "sentence2": "the coherence scheme only needs to keep track of a small number of logical sharers that get mapped into the subset of phys­ical cores used by the application.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2198", "sentence1": "Based on the above implementation details, we see that SR and HR have a similar software and hardware flow.", "sentence2": "sharer domains and home domains can be managed similarly.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2199", "sentence1": "A smaller line size results in an increased number of evictions.", "sentence2": "canneal shows a higher number of clean evictions increasing CET .", "label": "entailment", "domain": "Hardware"}
{"id": "test_2200", "sentence1": "Canneal, which shows high MPKI, benefits from using smaller size partitions in Phase 2, as frequent evictions lead to decrease in DET .", "sentence2": "there is a need for an accurate cache vulnerability modeling and a lightweight online estimation technique that accounts for both spatial and temporal vulnerabilities of concurrently executing applications under changing cache configurations.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2201", "sentence1": "we use the average value obtained during offline characterization of the application to estimate CCF.", "sentence2": "phase classification is based on clustering basic blocks considering their execution behavior and their frequency of occurrence.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2202", "sentence1": "Mix 6 also shows a similar behavior.", "sentence2": "the observations described above led us to explore the possibility of reconfiguring the cache to increase its reliability (i.e. reduce its vulnerability) while running different applications.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2203", "sentence1": "Fingerprint-based comparison has better performance but needs to resolve potential collision problems, compared to byte-tobyte comparison.", "sentence2": "sHA-1 hashing is one of the most commonly used methods in generating fingerprints of selected data because SHA-1 has an extremely low collision rate and, at the same time, reasonably good performance [10, 16, 20].", "label": "entailment", "domain": "Hardware"}
{"id": "test_2204", "sentence1": "The objective is to optimize the PCM space utilization with a good balance in the system performance and the space utilization, where data deduplication is adopted to increase the de facto capacity of a storage device.", "sentence2": "a containerbased space manager is proposed to increase the efficiency of the PCM space utilization as more and more data are written.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2205", "sentence1": "To address this issue, we propose to adopt wear leveling for containers with the integration of the proposed two-class bucket list.", "sentence2": "when a container is selected by procedure C ONTAINE R -ALLO C (in Algorithm 3), this container is compared with the first container of each bucket in both the compact and non-compact lists by checking their age .eld in the their container metadata records.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2206", "sentence1": "In order to minimize the frequency on allocating containers from the proposed bucket list and to ultimately extend the time on repacking chunks of containers, we adopt the worst-fit policy that tries to allocate containers with the largest free space and prefers to allocate the free space without any internal fragments.", "sentence2": "the proposed strategy will scan the com­pact list from bucket N to bucket 0, and return the first container from the bucket that is not empty.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2207", "sentence1": "The rationale behind this design is that the storage device initially has more free space but less knowledge in data deduplication.", "sentence2": "there is no need to have too many compaction activities to optimize the space usage in the beginning.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2208", "sentence1": "What is the last mile of the challenges in the designs would be on how to effectively provide an operating system a PCM storage device with a common storage-device interface, which is usually of fixed management units such as an LBA (or referred to as a sector for hard disks).", "sentence2": "the technical problem falls on how to (1) efficiently manage (i.e., allocated and deallocate) free PCM storage space and (2) map fixed-sized LBAs to variable-sized chunks when CDC is supported, so that the space utilization of PCM-based storage could be ultimately improved.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2209", "sentence1": "Then the second field value 1024 indicates that LBA 2 starts with byte 1024 of chunk A.", "sentence2": "by deducting 1024 from 1100, we can derive that the last 76 bytes of chunk A are stored in the beginning of LBA 2.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2210", "sentence1": "In this way, the number of buckets can be significantly reduced to improve the performance on scanning the buckets for containers that have enough free space for new chunks.", "sentence2": "otherwise, the containers related to the returned tree node are merged as a larger container, which is then returned to store new chunks (Lines 4-5).", "label": "neutral", "domain": "Hardware"}
{"id": "test_2211", "sentence1": "Then, we removed a specific percentage of the generated data randomly and restarted the data generation pro­cedures.", "sentence2": "for the high duplicate work­load with large file size, such as fileserver, paging could achieve good performance.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2212", "sentence1": "This work is motivated by the demands of intelligent storage de­vices.", "sentence2": "we are interested in the designs of intelligent storage devices with the capability to store much more data than its original storage capacity.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2213", "sentence1": "In Figure 6 we assume without loss of generality that the execution time of cv1 at V-fmax is less than the timing constraint (denoted by deadline).", "sentence2": "let us consider that the power consumption at V-fmax exceeds the TDP constraint.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2214", "sentence1": "Figure 3 shows software vulnerability (in log scale), execution time (in terms of clock cycles) and overall reliability (see Eq.2) of different compiled versions for three applications.", "sentence2": "the power consumption at a high V-f level may exceed the TDP constraint of the underlying Dark Silicon chip.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2215", "sentence1": "In the rest of the algorithm, the main computation is performed to examine all V-f level scalings and code changes for all ready tasks and then putting them into a maxheap.", "sentence2": "the total power consumption of a core at a V-f level and threshold voltage Vth can be written as Eq.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2216", "sentence1": "All the stacks for a particular function share a single stack pointer.", "sentence2": "our technique considers exactly which variables are live across recursive calls and only stores those to a shared stack, which may lead to reduced memory usage.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2217", "sentence1": "We represent user-defined data types as statically-sized bit vectors in hardware; if a definition is recursive, we cannot determine how many bits to assign to the recursive component at compile time.", "sentence2": "we replace recursive portions of these data types with a pointer data type e.g. a non-empty integer list defined as Cons Int List becomes Cons Int Ref, where Ref can be defined as a primitive numeric type.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2218", "sentence1": "We select an energy efficient RF size for comparison against the nonoptimized and the pruned radix-2 implementations.", "sentence2": "if the vector register width is W , the number of memory  min(m, 2N k ) subset_acc(N, m, k, W ) = (7) W We use the ceiling operator so that if the number of non-zero points is not a multiple of SIMD width W , then the additional load fetches the remaining values.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2219", "sentence1": "For all experiments, our approaches and Daikon extracted the same set of invariants.", "sentence2": "from the accuracy point of view they are equivalent, while they differ from the execu­tion time point of view.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2220", "sentence1": "GPUs are multi-core coprocessors originally intended to speed-up computer graphics.", "sentence2": "their highly-parallel structure makes GPUs powerful devices also for the elaboration of general-purpose computing-intensive processes that work in parallel on large blocks of data.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2221", "sentence1": "Indeed, execution time of invariant mining depends on the number and length of the execution traces to be analysed, the number of considered variables, and the number of invariants actually present in the traces (due to the effect of the two optimizations described at the end of Section 4.1.", "sentence2": "experimental results have been conducted on randomly generated execution traces with different values for such parameters by considering boolean and numeric (integer and real) data-type variables.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2222", "sentence1": "Indeed, a T-invariant is a TW-invariant for a time window that extends from the first to the last simulation in­stant of the corresponding trace.", "sentence2": "to avoid burdening the discussion, in the following, we use the term invariant when concepts apply indistinctly to T-invariants and TW­invariants.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2223", "sentence1": "However, MD5 has other attributes that make it a useful complement to the various memorybandwidth and .oating-point intensive algorithms already in SHOC.", "sentence2": "like many hash functions, MD5 contains many integer and bit-wise operations.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2224", "sentence1": "This python code was used as a baseline to produce both a CUDA version, using CUBLAS and hand-­written CUDA, and an Intel Xeon Phi version, using OpenMP and offload primitives.", "sentence2": "when using the newer 14.6 version of the compiler, MD5Hash would run as expected, but performance of MaxFlops dropped to less than 50% of the device peak.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2225", "sentence1": "An experimentation procedure, described in detail in [15], has been designed to estimate all the required parameters that cannot be obtained analytically.", "sentence2": "[15] depicts how host and Xeon Phi power, execution time, and performance are estimated experimentally using application code instrumentation.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2226", "sentence1": "The model uses widely-used performance metrics, such as data movement and computational throughput, to estimate execution time and relies on static and dynamic power estimates to calculate the total power draw on each participating device.", "sentence2": "the proposed model may be used to access the power-capping affects for a given application.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2227", "sentence1": "Our compiler supports VLIW and SIMD vectorization.", "sentence2": "it is challenging for the programmer to make effective use of a complex MPSoC.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2228", "sentence1": "Reconstructing any specific design state is not considered more valuable than reconstructing any other state.", "sentence2": "practical debugging experience suggests that some signals are inherently more valuable for validation and de­bug than others.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2229", "sentence1": "Consequently, signals selected to optimize SRR do not necessarily facilitate debug.", "sentence2": "sRR is not useful for signal selection for designs with the following features.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2230", "sentence1": "However, this operation would overwrite the state of the flip-flops and the circuit cannot resume normal operation after it is completed.", "sentence2": "to maintain the state of the flip-flops, the bits that are being shifted out from the scan chain are fed back into it through a multiplexer, as shown in Fig. 4.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2231", "sentence1": "In [24], the design of a Texas Instruments 3.5G baseband and multimedia applications processor is pre­sented.", "sentence2": "increasing the number of scan­chains increases the size of the MISR, and therefore, more area is required to implement it.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2232", "sentence1": "The microcontroller can read and write to the registers on every on-chip core through this interface.", "sentence2": "the signature and the sensor data can be accessed by the microcontroller through the I2C interface using buffer read operations.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2233", "sentence1": "CHStone suite is a set of 12 benchmarks, explicitly collected for the evaluation of High Level Synthesis flows, which aim at representing all the possible scenarios which have to be addressed by an High Level Synthesis tool.", "sentence2": "this suite contains both data dominated applications (aes, blow.sh, jpeg, mpeg2, sha) and control oriented applications (adpcm, dfadd, dfdiv, dfmul, dfsin, gsm, mips).", "label": "entailment", "domain": "Hardware"}
{"id": "test_2234", "sentence1": "In particular 1 obtains better results on mips thanks to the implementation of Rotation Scheduling.", "sentence2": "3 obtains better results on dfdiv and dfsin thanks to the implementation of a better division algorithm.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2235", "sentence1": "For example, if we use dedicated interconnections, we can connect a given F Ui to at most max_fanin + max_fanout other FUs.", "sentence2": "due to the recent high transistor scaling rate, reached with the use of new technologies [1] [2] [3], the transistors density per area unit is hugely increased.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2236", "sentence1": "However, we need to take into account the non-conflict constraint on a given buslet: for this, we need to ensure that at most one FU per clock cycle will drive the buslet (this will be done via tri-state buffers, enabled at the right clock cycle(s) by the datapath FSM controller) In order to ensure this, a proper scheduling algorithm is needed which minimizes the number of buslets required for such a constraint; a corresponding binding algorithm with other minimization goals and constraints is also needed.", "sentence2": "our algorithms solve a variation of the MR-LCS problem that we call MRW-LCS (minimum resources and wiring, latency constrained scheduling), and is divided into two stages: 1) Scheduling algorithm, which is itself divided into two phases: a) FDCS (Force Directed Communication Scheduling), a modified version of Force-Directed Scheduling [6], minimizes the lower-bound number of buslets implied by scheduling communications between operation nodes (the number of buslets could increase from this lower bound during binding, as we need to satisfy some extra constraints there).", "label": "entailment", "domain": "Hardware"}
{"id": "test_2237", "sentence1": "Dashed arrows indicate communication requested between resources, a different color indicates a different clock cycle in which a communication is scheduled; while thick lines indicate wires.", "sentence2": "in the next section, the concept of flexible buslets and their potential bene.ts to VLSI designs will be discussed.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2238", "sentence1": "Here, delay bounds on paths are treated in the same way as routing space constraints.", "sentence2": "delay violations are not banned but minimized simultaneously with routing congestion.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2239", "sentence1": "To calculate the down stream capacitance for each si, we traverse the net tree from sinks to source in a bottom-up manner.", "sentence2": "the down stream capacitance of the source segment, i.e. the segment connected with the driver pin, should be calculated after all the other segments have obtained their down stream capacitances.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2240", "sentence1": "Although top metal layers are less resistive than those in lower (thin) metals, it is impossible to assign all wires to top layers.", "sentence2": "layer assignment should satisfy the capacity constraints on thick metals.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2241", "sentence1": "Since timing requirements within a single net are usually different for different sinks, assigning all segments of a set nets on higher metal layers is not the best use of critical metal layer resources.", "sentence2": "intelligent layer assignment should not blindly assign all segments of a net to a set (a pair, for example) of higher metal layers.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2242", "sentence1": "Unlike the clock tree with links between clock nodes, which is a sort of an incremental modification of the structure of clock tree, clock spine network is a completely separated structure from the structures of tree and mesh. ", "sentence2": "it is necessary and essential to develop a synthesis algorithm for clock spines, which will be compatible to the existing synthesis algorithms of clock trees and clock meshes.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2243", "sentence1": "Since the specification items are tightly related each other and it is practically very time consuming to explore all feasible clock spine candidates, we solve the problem of clock spine synthesis by performing four steps.", "sentence2": "our proposed power-aware clock spine synthesis called CSPINE consists of four steps: (Step 1) Clustering clock sinks, (Step 2) Relaxing spines, (Step 3) Inserting spine buffers, and (Step 4) Constructing a top-level clock tree.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2244", "sentence1": "(Note that the previous power-aware clock tree synthesis algorithms (e.g., [13]) performed SPICE level simulation.", "sentence2": "they considered skew values only.)", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2245", "sentence1": "The results are stored into a on-chip non-volatile memory.", "sentence2": "physical unclonable function (PUF), first introduced by Gassend et al. [5] based on inevitable manufacturing variations, can be used to generate unpredictable and unclonable IDs. ", "label": "neutral", "domain": "Hardware"}
{"id": "test_2246", "sentence1": "As reported by US Department of Commerce [1], counterfeit­ing accounts for more than 8% of global merchandise trade, which is in the trillion dollars range.", "sentence2": "counter­feit electronics are increasingly being distributed throughout the market and legitimate electronic companies miss out on about $100 billion of global revenue every year because of counterfeiting, thereby posing a severe threat to the global electronic supply chain [2].", "label": "entailment", "domain": "Hardware"}
{"id": "test_2247", "sentence1": "Various techniques have been presented in the literature to generate unique IDs for IC products [4].", "sentence2": "physical unclonable function (PUF), first introduced by Gassend et al. [5] based on inevitable manufacturing variations, can be used to generate unpredictable and unclonable IDs. ", "label": "entailment", "domain": "Hardware"}
{"id": "test_2248", "sentence1": "The ARFIMA model allows for incorporating the long term memory of the process x(t) by capturing the fractality within its fractional differencing operator.", "sentence2": "for instance, for a large scale NoCbased multicore platform capable of both wormhole packet and network coding [11] communication, the MPC-based optimization modules have to decide when to switch between the two approaches such that the power/energy consumption profile is minimized while not sacrificing overall system performance.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2249", "sentence1": "The length N of the SNs used in a stochastic computation controls the accuracy and the total energy consumed by the circuit.", "sentence2": "by decreasing N, one can trade away accuracy for energy or power savings.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2250", "sentence1": "Our approach to this problem is a straightforward search within the operating-point space.", "sentence2": "we will try different operating points and, for each, evaluate the accuracy and energy of the corresponding circuit.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2251", "sentence1": "While CSE preserves the cluster structure through the conversion from a proximity representation to a Euclidean vector representation (e.g., the cluster structure is preserved between E1 and P1 , between E2 and P2, and so on), the k-means cost function in the original feature space O1 would also be identical to the cost function of pairwise clustering in the proximity matrix derived using Euclidean distance P1 [19].", "sentence2": "applying k-means clustering in E1 is equivalent to applying k-means clustering in O1.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2252", "sentence1": "From the high-dimensional base feature space FM .FB .FN , the proximities are able to reveal the underlying horizontal stripe pattern on the wafer even though in most test items this pattern are not directly observable and shadowed by some other more dominant types of spatial patterns.", "sentence2": "such stripe spatial variation may be subtle but consistently exists in most of the test items.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2253", "sentence1": "In fact, the classification accuracy based on the base features plus the two kPCA ­based features (LOF value and the first dimension of the embedded space) never surpasses the classification accuracy based on the base features only, in the range we searched for an optimal pair of SVM parameters [18].", "sentence2": "including both subsets of the proximity based features for classification could lead to a significantly greater test escape detection rate than including either of the subsets alone.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2254", "sentence1": "The idea is to generate an emulated test escape pool such that the corresponding test items of those hidden failing measurements are widely spread among a large number of test items and no single test item can directly detect a large number of emulated test escapes.", "sentence2": "we identified test items that hiding each of them would cause an increase of test escape rate by more than 50PPM to test escapes, and removed the test escapes created by hiding these test items from the emulated test escape pool.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2255", "sentence1": "CSE involves eigendecomposition of the proximity matrix [14].", "sentence2": "the embedded space is composed of the eigenvectors of the proximity matrix.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2256", "sentence1": "In uniform TSV placement, f-TSVs are placed uniformly, and they are considered as placement obstacles when logic cells (IP blocks) are placed in the 3D placement stage.", "sentence2": "in non-uniform TSV placement, the f-TSVs are added to the 3D netlist as TSV cells, and then placed with the logic cells (IP blocks) simultaneously during 3D placement.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2257", "sentence1": "In order to boost the likelihood of successful repair when multiple TSVs are faulty in the same group, it is bene.cial to increase Ngs , but it also leads to higher hardware cost.", "sentence2": "it is important to explore the trade-off analysis between TSV yield and hardware cost.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2258", "sentence1": "For s-TSVs, as discussed in Sec­tion III, they can only be inserted in whitespaces.", "sentence2": "a branchand-bound based approach is used to extract the whitespace [34], and then the location candidates for the s-TSVs can be determined based on the amount of whitespace and the area of a single TSV.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2259", "sentence1": "Clustering leads to the locationdependency of TSV defect probabilities [14].", "sentence2": "the defect probability of each TSV is related to its location.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2260", "sentence1": "For the uniform case, since the locations of f-TSVs are determined before logic cells are placed, s-TSVs and the supporting infrastructure (i.e., MUXes and wires) can be inserted right after f-TSV planning but prior to the placement of the logic cells and detailed routing [15].", "sentence2": "it is not necessary to consider the whitespace constraint during s-TSV allocation.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2261", "sentence1": "Then, due to location dependency, the defect prob­ability of the relocated f-TSV will change after the allocation of the s-TSVs, which can result in a situation where repair is not feasible.", "sentence2": "the whitespace constraint must be taken into account during s-TSV allocation.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2262", "sentence1": "Redirection and masquerade prevention: If A sends a message m to B, then the message must be delivered to B.", "sentence2": "it should be impossible for a (potentially rogue) IP C to masquerade as B, or for the message to be redirected to a different IP D in addition to, or instead of B.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2263", "sentence1": "To understand how this can be done, note that each router includes a base address register (BAR) which is used to route messages for specific destinations.", "sentence2": "one of the routers in the proposed path, Router0 is connected to the CPU; the BARs in this router are subject to potential overwrite by the host operating system, which can consequently redirect a message passing through Router0 to a different destination.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2264", "sentence1": "This architecture is specifically targeted towards smartphones and provides secure separation features to enable information partition between business and personal content coexisting on the same system.", "sentence2": "it permits hot swap between these two content worlds (e.g., without requiring system restart).", "label": "entailment", "domain": "Hardware"}
{"id": "test_2265", "sentence1": "Unfortunately, the state of the practice in this area by far is extremely manual and depends heavily on human creativity and observation.", "sentence2": "a security architect roughly iterates through the following five steps until convergence.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2266", "sentence1": "From another side, some HLS tools generate the SystemC simulation model after synthesis is done.", "sentence2": "we can use the powerful loop transformation in PolyOpt to expose atomic tiles in the source code and transfer them to SystemC modules.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2267", "sentence1": "In practice, TLM is supported by SystemC where the computation and communication models are implemented as function calls and channels.", "sentence2": "the creation of such a SystemC model is not easy.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2268", "sentence1": "Thus, the latency is highly predictable at the SystemC-level.", "sentence2": "to achieve diversity, we choose benchmarks from computation kernels with different applications.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2269", "sentence1": "Second, most of these studies stay at RTL level and the accuracy of this method is limited when applied at high level (also known as behavior level), where little hardware details is available.", "sentence2": "some assumptions and predictions must be made on the hardware implementation, which limits the accuracy.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2270", "sentence1": "Despite the significant speedup in simulation time, it is computationally infeasible to exhaustively simulate every design with simulation.", "sentence2": "it is pragmatic to apply a technique, such as regression modeling, to characterize the design space with a reduced sample of fast simulations.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2271", "sentence1": "SynchroTrace requires generating traces once for each thread count (i.e. a trace for each 2, 4, 8, 16, 32, and 64 threads per benchmark) and has a total compute time on the order of 100 years.", "sentence2": "the area of these design configurations is calculated with Cacti 6.5 [22] for the caches and Orion 2.0 [23] for the NoC using the 65nm technology.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2272", "sentence1": "Overall, as shown in Table II, the proposed regression modeling methodology enables a .ltered optimal design space, on the order of 100s to 1000s of design configurations, to be discovered out of the total 2.4 million design configurations.", "sentence2": "the regression models constructed with uniformat-random sampling all have a systematic bias.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2273", "sentence1": "IA cores are simulated at 2.0 GHz.", "sentence2": "the simulation is partitioned into multiple threads across the irregular line a Simics thread representing the base platform with the IA core, uncore, system memory, Peripheral Control Hub (PCH), network interface and the South bridge components needed to boot the platform, and one or more SystemC threads each running a SystemC device model as part of the VP.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2274", "sentence1": "Fifth, we demonstrate how the visualization tool can be used to assist architects to better design the system by a case study for the NUCA design.", "sentence2": "the GAM maintains a resource table to track the available hardware accelerators (or accelerator building blocks, i.e., ABBs), and the waiting time of those that are currently in use.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2275", "sentence1": "To make a TMG (pipeline network) deadlock-free, all empty cycles must be broken.", "sentence2": "since the TMG is derived from the original pipeline network, we cannot simply add tokens to empty cycles.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2276", "sentence1": "Both methods can pull cells toward their corresponding regions.", "sentence2": "when cells are already in their corresponding regions, only the method with two anchors can avoid further interference from pseudo nets.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2277", "sentence1": "Moreover, an initial solution without considering fence region constraints might lead to longer convergence time in the following analytical global placement stage.", "sentence2": "we propose a two-round quadratic placement algorithm to spread fence cells to the corresponding regions more closely by the following three steps: (1) we first apply a first-round QP to obtain a solution (see Figure 3(a)).", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2278", "sentence1": "Therefore, a cell assigned to this fence region tends to move toward the center even it is already in this region, which is not desirable.", "sentence2": "rather than adopting an anchor assignment technique, we propose a new density model which considers fence region constraints and minimizes cell overlaps simultaneously.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2279", "sentence1": "Divide and con­quer used to be applied in global placement, such as partitioning based methods [4, 5, 7, 29 31].", "sentence2": "since analytical techniques were proposed, partitioning based approaches are considered inferior.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2280", "sentence1": "Each sub-placement would use at most two threads, for example, when generating (or solving) linear systems (3) and (4) simultaneously.", "sentence2": "pOLAR 3.0 launches at most 30 threads.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2281", "sentence1": "In such scenario, timing estimates that consider only logic synthesis were accurate enough.", "sentence2": "in technologies below 90nm the delays of interconnections became much more significant as compared to the delays of logic gates [9] [22].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2282", "sentence1": "Observe how our technique drastically reduces the electrical effort (up to -9) of some critical gates that were overloaded before our optimization.", "sentence2": "it is possible to conclude that such unloading helps to improve both WNS and TNS since a critical gate may influence several other near-critical ones.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2283", "sentence1": "The security wrappers provide a standardized way for E-IIPS to obtain such collateral while abstracting the details of internal implementation of individual IPs.", "sentence2": "the wrappers implement a protocol to communicate with E-IIPS during the execution.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2284", "sentence1": "Unauthorized or malicious access to these assets can result in leakage of company trade secrets for device manufacturers or content providers, identity theft for end users, and even destruction of human life.", "sentence2": "it is vital to ensure that secure assets in computing devices are adequately protected.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2285", "sentence1": "This allows us to say that for a greater value of the mean mutual information we will see a greater mean success rate indicating greater information leakage.", "sentence2": "both success rate and mutual information are able to describe timing leakage for cryptographic hardware architectures.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2286", "sentence1": "That is due to the fact that architecture stops its execution after it reaches the most significant 1 bit of the key.", "sentence2": "we can determine with great accuracy where this one 1 bit resides due to the overall runtime of the algorithm.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2287", "sentence1": "This shows how much information leaks from the i-th key bit when encryption time is known.", "sentence2": "how much does the total runtime depend on the i-th bit of the key", "label": "entailment", "domain": "Hardware"}
{"id": "test_2288", "sentence1": "We assume that the attacker can determine the amount of cycles for each execution of the hardware component.", "sentence2": "abstract Cryptographic function implementations are known to leak information about private keys through timing information.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2289", "sentence1": "We assume that the attacker can determine the amount of cycles for each execution of the hardware component.", "sentence2": "the attacker can determine the cycle in which the execution begins and ends.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2290", "sentence1": "In the conventional method, the sample number N must be larger than 2FtTg to avoid aliasing of gd[n].", "sentence2": "with the constraint in (41), our method requires much less samples of the channel's frequency response than the conventional method.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2291", "sentence1": "Most typically, the number of chemical species and reactions are much smaller than the number of possible states and transitions between these states.", "sentence2": "in the stochastic chemical kinetics formalism, both the sparsity of possible transitions between the states, and more importantly, the special structure of these transitions caused by a small set of reactions, are exploited to arrive at a very compact representation for a MC that may be otherwise very unwieldy to deal with.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2292", "sentence1": "Most often, a well-stirred system is assumed so that concentration of molecules of each type is uniform over the entire spatial domain [38].", "sentence2": "instead of tracking the spatial positions of each and every molecule in the system, the system state is compactly represented by a vector containing the number of molecules of each type [21].", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2293", "sentence1": "Nevertheless, if the generic MC neither has a sparse state transition diagram nor possesses any special structure, using such a formalism does not result in a more compact representation compared with the state transition diagram.", "sentence2": "for a generic MC with no special structure, the required number of abstract species and reactions will be equal to the number of states and transitions.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2294", "sentence1": "The intricacies in the numerical treatment of (2) arise due to the square root expression in the time-dependent diffusion part of the SDE.", "sentence2": "the possibility of the square root becoming imaginary needs to be addressed.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2295", "sentence1": "Our merging is more powerful than BRMap’s packing [8], since BRMap only packs operations from if-path and elsepath of ITE guarded by an identical predicate.", "sentence2": "operations in the loop kernel updating the same variable may be predicated by various of combinations of predicates when confront nested ITE (NITE) (see figure 6). ", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2296", "sentence1": "With such capability, the proposed design framework can potentially be used for increasing design productivity for any heterogeneously integrated CMOS-NEM design (e.g.", "sentence2": "design capture is based on a NEM relaybased standard-cell library.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2297", "sentence1": "It varies depending on how soon the relay is reactuated, because a de-actuated relay beam oscillates until the stored potential energy of the beam is dissipated through damping.", "sentence2": "the switching-on time is dependent on the position and velocity of the beam at the time of actuation.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2298", "sentence1": "Inter-die coupling E-.elds are mostly limited between a few metal layers because of E-field shielding from metal wires.", "sentence2": "it is safe to ignore a few metal layers as in our in-context extraction, which still captures most of inter-die coupling E-.elds.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2299", "sentence1": "Note that though inter-die coupling capacitance is a large portion of total coupling capacitance, ground capacitance and pin capacitance are major contributors to the load of a net.", "sentence2": "inter-die coupling only affects slightly on the switching power consumption of F2F designs.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2300", "sentence1": "Overall, total capacitance always increases with a closer die­to-die distance, and the portion of inter-die coupling keeps increasing as well.", "sentence2": "die-by-die extraction, which is unaware of the neighbouring die and ignores the E-field sharing, cannot extract the inter-die coupling capacitance accurately.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2301", "sentence1": "As long as the total weight of both dies is equal to one, there is no overestimation in inter-die coupling.", "sentence2": "for a double-counted capacitor, the weight ratio between the bottom in-context die and the top in-context die is proportional to the product of R ratios of both layers the capacitor connects to.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2302", "sentence1": "For example, it can be seen in Fig. 5 that for the gate nets of transistors M1 and M4, the conditions for the unswitchability property for p edges are fulfilled.", "sentence2": "the p edges belonging to M1 and M4 are unswitchable.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2303", "sentence1": "In this work, we are going one step beyond verification to an automatic, i.e., fault free synthesis of power-down circuitry.", "sentence2": "this would introduce head or foot switches which disconnect the ground and/or supply from the circuit.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2304", "sentence1": "For each rip-up path, it has to be decided where it should be ripped up following the objectives (B)-(E).", "sentence2": "the following algorithm was implemented: In order to achieve a rip-up of current mirrors like it would be done in a handcrafted design (see Fig. 6), current mirrors are detected by a structure recognition method [14], [15], [11].", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2305", "sentence1": "When receiving the response, NIC inspects the SSV of the response to check if the response is sent from the latest data owner.", "sentence2": "the response is confirmed to be valid if there is no preceding write request for the cache line which is not snooped in the SSV from the response(Accordingly, the SSV only needs to provide the snooping status for preceding requests.).", "label": "entailment", "domain": "Hardware"}
{"id": "test_2306", "sentence1": "Outgoing coherence request from the local processor is kept at the outgoing request buffer (oRB) until it receives the data response and completes coherence operation.", "sentence2": "the response is confirmed to be valid if there is no preceding write request for the cache line which is not snooped in the SSV from the response(Accordingly, the SSV only needs to provide the snooping status for preceding requests.).", "label": "neutral", "domain": "Hardware"}
{"id": "test_2307", "sentence1": "NIC issues notifications only at the start of each time window, which is set greater than the maximum broadcast latency.", "sentence2": "all nodes receive the same set of notifications at the end of each time window.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2308", "sentence1": "Further, if two or more processors attempt to write the same memory location simultaneously, processors might observe write values in different orders from each other.", "sentence2": "we need well-defined rules specifying correct shared memory behavior so as to provide a basis in writing parallel programs.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2309", "sentence1": "Cache coherence defines memory access ordering for the same memory location by two constraints: (1) write must be eventually seen by other processors; (2) writes to the same location must be seen in the same order by all processors.", "sentence2": "to show the generality of our proposed Ordered-NoC\nplatform, we applied it to MPSoC designs running atop two different topologies (2D-mesh and butterfly fat tree) and supporting two alternative memory models, total store order(TSO) and relaxed memory consistency.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2310", "sentence1": "The STAGE algorithm can benefit if we can specify the starting state distribution using some domain knowledge.", "sentence2": "we also consider a starting-state distribution named a-Greedy.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2311", "sentence1": "In this work, we advocate that the concept of small-worldness should be adopted in 3D NoCs too.", "sentence2": "the vertical links in 3D NoC should enable the design of long-range shortcuts necessary for a SW network.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2312", "sentence1": "To compensate for the loss in performance due to TSV failure, fault tolerant router and NoC architectures with redundant vertical links [10] were proposed.", "sentence2": "these designs give rise to additional area and power overheads.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2313", "sentence1": "This problem is generally addressed by adding a penalty term to discourage complex functions.", "sentence2": "efficient optimization can overcome such limitations.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2314", "sentence1": "Apparently, the frequency resolution of such a representation is fmax/n.", "sentence2": "in many IPs, it is common that the wire cap is more than the device parasitics cap, which has been silicon validated on on-market products [22].", "label": "neutral", "domain": "Hardware"}
{"id": "test_2315", "sentence1": "It needs to be clarified that since entropy production has only been experimentally proved on metals, our fatigue analysis only focuses on the TSV body and the attached landing pad.", "sentence2": "it is limited to copper.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2316", "sentence1": "In the simulations, we also discover that the rate of entropy production increase in the landing pad and the TSV body through cycles remain the same.", "sentence2": "single cycle modeling is sufficient for TSV fatigue analysis using entropy production, which will be used in the remaining of the paper.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2317", "sentence1": "According to the second law of the thermodynamics, entropy is generated when the permanent degradation is presented through an irreversible process, which leads to a monotonic increase of disorder in a system.", "sentence2": "the component of material degradation can be measured by entropy and thermodynamics energy [13], [14].", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2318", "sentence1": "In the experiment, the chain has 50 links repeatedly, alternating between M2 and MC, which are local interconnects linking to other stacks.", "sentence2": "this unsymmetrical structure would lead to a non-homogeneous stress on vias.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2319", "sentence1": "Furthermore, by applying the entropy degradation theory related thermodynamic forces, Amiri [16] experimentally determined the critical damage value of specimens subjected to different operations based on the entropy flow.", "sentence2": "this method has only been applied to large scale analysis (in scale of meters), not to micro structure.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2320", "sentence1": "In F2B bonding, TSVs are used for vertical interconnects.", "sentence2": "since TSVs penetrate through the silicon substrate and occupy area, using excessive TSVs lead to area overhead, which designers should avoid.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2321", "sentence1": "Block-folding technique helps to reduce significant power in three-tier design.", "sentence2": "in this effort, three-dimensional integrated circuits (3D ICs) using through-silicon vias (TSVs) have gained a great deal of attention as a viable solution for low-power IC designs.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2322", "sentence1": "SEU behavior on logic resources, e.g. LUTs, has been well discussed in [1].", "sentence2": "the m inputs of each multiplexor are individually controlled in a fully decoded style [16] and the n multiplexors are sharing the same m con.guration bits.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2323", "sentence1": "In the synchronous execution model, data from the previous iteration is accessed by the neighbors.", "sentence2": "asynchronous model allows access to the latest data available.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2324", "sentence1": "Methods such as [7], [8] skip over the noncompliant stimuli by re-seeding the LFSR to a functionally-compliant value.", "sentence2": "the methods from [12], [13] let the LFSR run autonomously, however the stimuli are corrected in real-time using a set of encoded cubes placed on-chip.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2325", "sentence1": "In particular, by setting WX = m, the algorithm will select cubes with large implied space first strategy.", "sentence2": "it collects the cubes with the largest implied space, and selects one with the lowest correlation.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2326", "sentence1": "The addressing logic can record the status of cubes and overwrite the used cubes by new cubes that can be uploaded after all the patterns from the initial cube set have been exhausted.", "sentence2": "the cube RAM serves as a fist-in-first-out cube queue and refreshes its content as the validation process progresses.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2327", "sentence1": "However, a fixed length LFSR as used in [12], [13] lacks the flexibility of varying distribution of the sequences according to cubes, which causes stimuli repetition.", "sentence2": "a dynamic LFSR is designed based on it, and we select ξ (defined in Equation 2 from Section III) to control the behaviour of the dynamic LFSR as a degenerated ξ-bit LFSR, as detailed next.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2328", "sentence1": "First, on the one IP block that got full coverage, it happened at the same time that all instrumented CMs were covered.", "sentence2": "full coverage was only achieved when all instrumented CMs were set.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2329", "sentence1": "we found that even the naaive method provided good guidance in prioritizing which CMs are better to monitor.", "sentence2": "the fatal flaw with the native method is that it can severely miscompute the probability of full coverage: under the assumption of independence, Eq.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2330", "sentence1": "For the na¨ive method, there are no scalability issues, and the selection algorithm is simply to pick the hardest-to-cover CMs greedily.", "sentence2": "as noted earlier, the na¨ive method miscalculates the probability of full coverage, so if we wanted a probability of greater than 98%, the na¨ive computation often suggests including all or almost all CMs.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2331", "sentence1": "non-uniform cooling that increases on-chip thermal gradients giving rise to reliability issues and shorter lifetimes.", "sentence2": "it is fundamental to optimize several parameters, such as the pressure drop across the channels, the number of fluid inlet and outlet ports, and the fluid network, to achieve optimized cooling in terms of energy efficiency, thermal gradient and peak temperature.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2332", "sentence1": "In the former, improving the slacks of a subset of paths does not guarantee an improvement on total negative slack (TNS) or worst negative slack (WNS).", "sentence2": "improving the slack on one failing path can cause several originally-passing paths to now fail.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2333", "sentence1": "As the logic function for cells was not known in the original ICCAD-2012 contest benchmarks, the methodology in [15] created a (unverified) mapping to standard cells based on their sizes and pin counts.", "sentence2": "a given cell would be mapped to a cell type that has the same input/output pin counts and the same or slightly smaller width in the given cell library.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2334", "sentence1": "FHT also removes duplicated and circulated traffic from the rings.", "sentence2": "the FHT approach significantly reduces redundant unicast traffic in HSR networks.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2335", "sentence1": "Traffic-filtering based techniques mainly focus on solving the HSR s issues.", "sentence2": "these techniques try to filter unicast traffic for unused rings in HSR networks and remove duplicated and circulated traffic from the rings.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2336", "sentence1": "FPGA-based image processing can run parallel calculations and stream processing.", "sentence2": "fPGA-based image processing has latency, not delay [11] and can operate faster than PC-based processes.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2337", "sentence1": "If the reference count is more than two, proposed algorithm regards the page as a hot page and the page is migrated to the MRU position of DRAM [Figure 2d].", "sentence2": "clean pages on PCM are not migrated to DRAM even if its reference count is more than two because read latency of PCM is similar to that of DRAM [Figure 2e].", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2338", "sentence1": "Especially, our algorithm performs better in the range of small size memory.", "sentence2": "our algorithm is suitable for the process which issues a lot of read/write requests at critical state.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2339", "sentence1": "LMK does not kill the process whose oom_adj value is below zero such as system process and foreground process.", "sentence2": "the pages, which are accessed by non-critical processes, are firstly placed on PCM because non-critical processes do not need fast response time.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2340", "sentence1": "These two types of processes can affect to the user directly or indirectly because they are visible to user.", "sentence2": "algorithms need considerably long time to run the same workload because they do not give higher priority to critical process.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2341", "sentence1": "We present EnhancedTouch, a novel bracelet-type wearable device for facilitating human-human physical touch.", "sentence2": "we aim to support children with autism spectrum disorder (ASD), who often exhibit particular communication patterns, such as lack of physical touch.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2342", "sentence1": "A possible solution is to employ electric field PAN technology, where the electric field induced around the surface of human body is used as a communication channel and the electrodes do not have to attach to the skin.", "sentence2": "in this case, accurately differentiating the touch and proximity is a challenging task because RSSI depends not only on the contact states but also on the spatial relationship between the devices.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2343", "sentence1": "In such scenarios, a particular CL still follows the most suitable hierarchy at run-time.", "sentence2": "a CL follows a two-level hierarchy when there are multiple copies of itself; the same CL follows a three level hierarchy when there is only a single-copy in the system, thus optimizing the access in both cases.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2344", "sentence1": "To this end, the Simulink model of the control design is updated with extra delay blocks.\n", "sentence2": "a parametrised rule-based model transformation is created as depicted in Figure 3.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2345", "sentence1": "For example, at time stamp 8 s the driver issues a lower-command resulting in an ideal reaction time of 52 ms.", "sentence2": "after deployment the reaction time appears to be 221 ms which is 10 % higher than required.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2346", "sentence1": "During these successive iterations, the available ADL can be used to enhance the deployment process.", "sentence2": "the Design-Space Exploration (DSE) process for optimal deploy­ment can be influenced by taking into account information available in the ADL.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2347", "sentence1": "The GAP algorithm that is used in the CALIC design, also tries to take several different edge types into account.", "sentence2": "it accumulates all absolute vertical changes and compares it with the accumulation of all the absolute horizontal changes.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2348", "sentence1": "For the red and blue color, this is sufficient.", "sentence2": "in this paper, it is proven that a synergy can be reached by combining a color filter array (e.g. Bayer pattern) with known predictive-corrective coding filters and an entropy encoding for a higher overall lossless compression of images.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2349", "sentence1": "To be correct, it must be mentioned that for this research the GBRG Bayer pattern (as in figure 6) has been chosen and reverted from the benchmark images.", "sentence2": "the results only give a good indication, but are not the most correct values.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2350", "sentence1": "Also, all the image details need to be preserved for improving the computational usage in a later stage.", "sentence2": "this research is focused on predictive-corrective coding filters with entropy encoding (i.e. Huffman coding) and apply these on the raw image sensor data to compress the huge amount of data in a lossless manner.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2351", "sentence1": "As the image sensor will transmit the 2D image in a pixel serial order (i.e. left to right for each row and from top to bottom), a prediction can only be done for the next pixel in line.", "sentence2": "none of the future information (i.e. on the right and below the current pixel) can be used, unless there is a framebuffer.\n\n", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2352", "sentence1": "Meanwhile, the LEA will encrypt all candidate jmp destinations and triggers jmp decryption while executing jmp instructions, but some instructions at candidate destination addresses of jmp may also be executed sequentially (e.g. do-while or if-statement alone) without being the des­tination of the jmp.", "sentence2": "recent research aim at balancing the tradeoff between security and such overheads [8-12] by revising the software or modifying processor’s instruction set architecture (ISA).", "label": "neutral", "domain": "Hardware"}
{"id": "test_2353", "sentence1": "The simulation results show that the false negative rate is less than 1%.", "sentence2": "attackers often overflow the stack to modify the return address and then chain the gadgets to launch a ROP at­tack.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2354", "sentence1": "The most general solution is control-flow integrity (CFI) [7] approach, where the control flow graph (CFG) [7] of the program is generated during compilation and is enforced at run-time [11].", "sentence2": "this has high memo­ry and performance overheads that make it impractical in many applications.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2355", "sentence1": "Dynamically updating the key would make the attack be harder since attackers must focus on each chip to launch an attack due to the uniqueness of PUF key for each chip, and even though the attackers can get the key luckily, but the key may be updated at regular intervals.", "sentence2": "a successful attack on a chip cannot be easily applied to another one due to the dynamic key.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2356", "sentence1": "The memory overhead is thus O(m).", "sentence2": "the first task of EDU is to encrypt return addresses and some bytes of the first instruction at jmp destinations. ", "label": "neutral", "domain": "Hardware"}
{"id": "test_2357", "sentence1": "In this work, we consider interconnects as groups of patterns belonging to different classes of frequently used layout pattern shapes.", "sentence2": "the full-chip reliability analysis can be carried out by independently evaluating the dielectric breakdown of interconnect patterns existed in the layout.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2358", "sentence1": "The x-­coordinates of the modules are well defined by the tree structure, starting from the root node n1 whose bottom-left coordinate is (0, 0).", "sentence2": "no published work achieves the optimal time complexity for general geometrical constraint handling and module packing.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2359", "sentence1": "CB-trees augment B*-trees with the corner stitching to provide a comprehensive system for analog place­ment problems.", "sentence2": "there are two deficiencies of CB-trees.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2360", "sentence1": "In addi­tion to neglecting the PBTI effects in nMOS transistors, the mathematical model itself is not able to consider the slope of rise and fall signals of a cell.", "sentence2": "this method fails to analyze multi-stage cells (e.g., buffers, flip-flops, etc.) and multi-gate paths.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2361", "sentence1": "Note that the obtained C P delay here is already in the presence of aging and hence no additional guardband needs to be applied.", "sentence2": "the included guardband can be com­puted as the timing difference between analyzing our design with the initial (i.e. degradation-unaware) and degradationaware libraries.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2362", "sentence1": "As illustrated in Section 2, the path that was initially (i.e. before aging) critical may not remain critical after aging.", "sentence2": "both aging-aware and aging-unaware designs will operate with no (required or contained ) guardband.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2363", "sentence1": "This work considers a 512-bit cache line consisting of eight 64-bit words.", "sentence2": "size of our meta-data (i.e., zero-flags and energy­-flags) is 16 bits (2 bits/word).", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2364", "sentence1": "In practical systems, N-point DFT is usually performed with the Fast Fourier Transformation (FFT) algorithm, which enables significant reduction in computing complexity from O(N2) to O(NlogN).", "sentence2": "from the perspective of hardware design, the computation-intensive FFT is still a resource-hungry block since an N-point FFT requires 0.5NlogN complex multiplications and NlogN complex additions, which cause very high hardware resource consumption in large N cases.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2365", "sentence1": "Figure 8: Comparison of .ne-grained (per set) and coarse-grained (global) rationing for symmetric co-runs.", "sentence2": "if resets are too infrequent, protection dom­inates sharing.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2366", "sentence1": "We have tested an extension of this policy for multiple cache-sharing programs.", "sentence2": "for each program, the determination of the hint bit is based on its ration size r.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2367", "sentence1": "They show that symp­toms are more prevalent in failed devices.", "sentence2": "a large fraction of healthy devices have no symptoms, and even if the symptoms occur, they are very mild in intensity, e.g., 80% of healthy devices have fewer than 2 reallocated sectors.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2368", "sentence1": "There is a higher likelihood of the symptoms (captured by SMART) preceding SSD failures, with an intense manifestation preventing their survivability beyond a few months.", "sentence2": "our analysis shows that these symptoms are not a sufficient indicator for diagnosing failures.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2369", "sentence1": "The reason is that industry is urgently exploring beyond-silicon and beyond-CMOS device, interconnect and memory options, as well as heterogeneous, More-than-Moore integration and packaging technologies, in order to maintain Moore s-Law scaling of integration value.", "sentence2": "silicon photonics stands out as the most promising technology to overcome the limitations of electronic interconnects as the system scale increases.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2370", "sentence1": "In this section, we evaluate the proposed method with the dataset collected from households with simple sensors and user-generated activity labels.", "sentence2": "we focus on the question whether the unknown activity classes are correctly estimated.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2371", "sentence1": "The off-line training cost is very high.", "sentence2": "they require a large number of training data of each previous program (e.g. Khan et al. [9], Guo et al. [12] and Dubach et al. [10, 11] require 1000, 500 and 512 training instances of each prior program, respectively).", "label": "entailment", "domain": "Hardware"}
{"id": "test_2372", "sentence1": "Inspired by the recent advances of transfer learning, we propose a transfer learning based DSE framework TrDSE to build a more efficient and effective predictive model for the target program with only a few simulations by borrowing knowledge from previous programs.", "sentence2": "trDSE includes two phases: 1) clustering the programs based on the proposed orthogonal array sampling and the distribution related features, and 2) with the guidance of clustering results, predicting the responses of configurations in design space of the target program by a transfer learning based regression algorithm.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2373", "sentence1": "TrDSE is much more efficient than PSM and EAC to achieve a similar prediction performance, as TrDSE only needs less than one fifth simulations of PSM and one third simulations of EAC.", "sentence2": "trDSE could dramatically reduce the simulation time in DSE.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2374", "sentence1": "Moreover, even in case a wearer can choose a size that perfectly fits, the pressure reduces over time and the original effect is lost.", "sentence2": "a pressure sensor and a measuring equipment that can correctly evaluate the pressure on each part of an individual's leg need to be developed for adequate use and investigation.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2375", "sentence1": "Each semester, full-time students received a $25 printing allotment and part-time students received a $15 printing allotment We used a networking tool provided by HP to notify the help desk of paper jams, low toner, paper outages and maintenance needs.", "sentence2": "printing was both free and unmetered printing was not tracked at all.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2376", "sentence1": "In [3], the authors use the EEG evoked potential to issue movement commands.", "sentence2": "the geometrical shape of the interface can be circular or rectangular, while the number of directions can be 4 or 8.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2377", "sentence1": "Furthermore, these sys­tems are not portable, they can only be used in particular spaces which the systems have been set up.", "sentence2": "these kind of motion capture systems are not suitable for daily activities monitoring.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2378", "sentence1": "Otherwise, the authors of [25] highlight the importance of efficient utilization of renewable energy sources within Cloud computing platform.", "sentence2": "to cope with the complexity, the authors introduce a fuzzy logic based on heuristic for GLB which improve the overall cost and the renewable energy utilization.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2379", "sentence1": "The fact of introducing the fuzzy control theory into resources scheduling, enables us to optimize the overall performance.", "sentence2": "this methodology makes it possible to reduce at the same time the response time, the processing time and the cost.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2380", "sentence1": "This business model has the particularity to offer unlimited capacity to final users.", "sentence2": "we highlight different categories of Cloud based on deployment models or service models (see Figure 1).", "label": "entailment", "domain": "Hardware"}
{"id": "test_2381", "sentence1": "These aspects are mainly, the security and the performance.", "sentence2": "performance optimization has been a strategic weapon in multiple scientific and industrial areas.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2382", "sentence1": "The maximization of this GPI allows us to achieve the best response time and processing at low cost.", "sentence2": "this approach has some limitations.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2383", "sentence1": "The scheduling methods in distributed systems aims to share the load on different processors and enhance their utilization while reducing the total task execution time.", "sentence2": "task scheduling is identified as combinatorial optimization problems which could increase the performance of the system.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2384", "sentence1": "For instance, it can hold a component firmly while the human is performing an assembly task.", "sentence2": "copyrights for third-party components of this work must be honored.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2385", "sentence1": "Hence the importance of Fracticality.", "sentence2": "validating personas in this manner became tricky, as interviews elicited powerful emotional responses from participants, necessitating caution.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2386", "sentence1": "These operations are essential to modern multi­media and signal processing engines, and form the basis of several industry standards [2].", "sentence2": "the scope of approximations in this experimental evaluation is restricted to floating point division only.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2387", "sentence1": "Architectures such as the proposed evalua­tion system can hence become essential for automatic performance analysis and style judging systems in future.", "sentence2": "a well-balanced data base for the subse­quent machine learning task could not be ensured.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2388", "sentence1": "In this work, we presented a human activity recognition sys­tem for wearable motion sensor data specialized in the error classification of ski jumping motions.", "sentence2": "we developed a multi-dimensional convo­lutional network model based on the idea of a three-­dimensional motion image connected along the temporal, skeletal and sensor domain.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2389", "sentence1": "Deep one-dimensional CNNs were demonstrated to outperform conventional classi.ers on sensor data from activity recognition data sets [20, 22].", "sentence2": "they are not known to have been compared to a multi-dimensional network as proposed in this work.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2390", "sentence1": "Moreover, it offers the chance to considerably enhance the credibility of judged sports known to suffer from biased evaluation [28]: to prevent controversies and allegations of subjectivity or even fraud, it is reasonable to include objective measures whose output cannot be manipulated in the final performance scores.", "sentence2": "the creation of autonomous machine intelli­gence that retrieves and classifies style information from an easily accessible stream of motion data appears an important future issue in affected sports.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2391", "sentence1": "On the other hand, the ball speed estimation shows slightly better results in comparison to existing methods like [15] or [16], which achieve an overall speed accuracy of 70 % to 85 %.", "sentence2": "for all results the standard deviations are clearly higher than the mean values, which indicates a high error distribution.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2392", "sentence1": "Since the friction coeffi­cient is always high enough, the ball will usually roll during the impact.", "sentence2": "the sliding case was not considered further.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2393", "sentence1": "Data was stored on an integrated memory and later processed for characterization of bowling deliveries.", "sentence2": "it is not possible to instrument a table tennis ball due to its very low weight.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2394", "sentence1": "For that, we needed a known resting point in the world frame, but only had inertial data in the local and moving racket frame.", "sentence2": "the intention of this approach is not to provide an exact description of racket movements resulting in accurate ball properties, but to proof that it is possible to apply different assumptions and simplifications for unknown occurrences to estimate the ball speed and spin shortly after impact.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2395", "sentence1": "In particular, transient HR increases caused by walking activities happen to be similar to 1st HR peak, and causes incorrect recognitions.", "sentence2": "heart rate data around walking duration is removed.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2396", "sentence1": "Preprocess Heart rate data is preprocessed to reduce temporal noises caused by various physical and psychological factors.", "sentence2": "transient HR increases caused by walking activities happen to be similar to 1st HR peak, and causes incorrect recognitions.", "label": "entailment", "domain": "Hardware"}
{"id": "test_2397", "sentence1": "Still, hardware design is also more challenging, simply because it is bound to interaction with physicality  which can change drastically during the lifetime of a system.", "sentence2": "it could have been the case that the transmission characteristics were rougher than in previous tests.", "label": "neutral", "domain": "Hardware"}
{"id": "test_2398", "sentence1": "Still, hardware design is also more challenging, simply because it is bound to interaction with physicality - which can change drastically during the lifetime of a system.", "sentence2": "a software system uses standardized interfaces to the world, which dramatically reduces possible problems and errors.", "label": "contrasting", "domain": "Hardware"}
{"id": "test_2399", "sentence1": "This proved to be difficult for the younger children, and expected to be even more so for children with compromised motor abilities.", "sentence2": "in the subsequent design we require only a single press to activate the recording, while its termination is done automatically by detecting silence longer than 500 milliseconds.", "label": "reasoning", "domain": "Hardware"}
{"id": "test_2400", "sentence1": "Hence it is interesting to understand what exactly can be enforced under this black box constraint.", "sentence2": "essentially EPxor (C) will now be a program that echoes inputs on outputs, and hence it is a secure program.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2401", "sentence1": "Edit automata are similar to our approach in that they offer a black box mechanism ca­pable of monitoring and/or editing the input/output behavior of pro­grams.", "sentence2": "to enforce infinite renewal properties, they have to buffer program actions  something that is undesirable for reactive programs.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2402", "sentence1": "Student feedback indicated that there were problems.", "sentence2": "student responses on a post-class survey revealed that students felt that the exercise was worthwhile and that it increased their interest in cybersecurity education (one-tailed t-test found ratings well above neutral p < .01).", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2403", "sentence1": "This workshop uncovered several issues with remote configuration and set up of the infrastructure, that were subsequently fixed.", "sentence2": "some exercises clearly involve networks and subnets.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2404", "sentence1": "Traffic of these clients can be closely monitored and forecasted.", "sentence2": "attack prediction module can be implemented at any of these locations -(i) only victim end (ii) only up­stream defense nodes (iii) both victim and upstream defense nodes by splitting the functionalities.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2405", "sentence1": "When a transmission is corrupted, the sender is aware of this event and it sends a rewind symbol ‘←’ on the next time it has the right to speaks.", "sentence2": "when the order of speaking is fixed (say, alternating), the parties “lose” one slot: while we would like the sender to repeat the transmission that was corrupted, the receiver is the next party to speak after the round where the ‘←’ symbol is sent.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2406", "sentence1": "So far, perfect profiling was assumed in the previous subsections.", "sentence2": "the computation is more time consuming and the solution cannot be written in a simple algebraic equation.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2407", "sentence1": "The definition does not demand that a security-mechanism actually does render the duplication more difficult, but only that an implemented mechanism is meant to do so.", "sentence2": "a classification of such architectures as two independent security and storage architectures is more appropriate.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2408", "sentence1": "For quantum-token based PUFs [11], the security objective D2 is reached via proving that the s-challenge is necessary to extract the previously stored information.", "sentence2": "this PUF prevents mathematical duplication with the help of the NC mechanism and cryptostorage with the EUR mech­anism.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2409", "sentence1": "In an attempt to construct two-pass leakage-resilient AKE protocols based on the eCK model, three models and pro­tocols have been proposed: The first formalization was the Moriyama-Okamoto (MO) model [8] which allows bounded amount of leakage.", "sentence2": "adversary: The adversary M is a probabilistic polyno­mial time (PPT) algorithm which controls all interaction and communication between parties.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2410", "sentence1": "For efficient data management and economic benefits, organiza­tions are increasingly moving towards the paradigm of database as a service\" by which their data are managed by a database man­agement system (DBMS) hosted in a public cloud.", "sentence2": "we observe that there is an overall in­crease by 44% on the server side.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2411", "sentence1": "By outsourcing data to the cloud, organizations save the cost of building and maintaining a private database system and have to pay only for the services they actually use.", "sentence2": "organizations are increasingly interested in the paradigm of \"database as a service\".", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2412", "sentence1": "Further, in such schemes achieving forward and backward security requires only to change the public information and does not affect the secrets given to existing subscribers.", "sentence2": "bGKM schemes do not support group membership policies over a set of attributes.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2413", "sentence1": " Samanthula et al. [21] propose an approach that relies on homomorphic encryption and PRE to handle user revocations without requiring re-encryption of relational data and also supports expressive policies.", "sentence2": "such approach incurs a heavy cost on the end user as all processing is done at the client side since the server does not perform secure query processing.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2414", "sentence1": "There are plenty of open sources towards how to use motion sensors to infer position of smartphones publicly available on many websites.", "sentence2": " attackers can easily get those code and add them into the repackaged app to make the stealthy picture taking triggered by motion sensors. ", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2415", "sentence1": "There are plenty of open sources towards how to use motion sensors to infer position of smartphones publicly available on many websites.", "sentence2": "for version 4.1.2 phones from 5 vendors, the success rate of transplantation attack is 75%.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2416", "sentence1": "They cannot be used outside their definition class.", "sentence2": "starting at the connect function does not work, and we should look for another start point.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2417", "sentence1": "This thread can know when and which API is called, and can get the caller s UID (user ID) and PID (process ID).", "sentence2": "through this thread, Android API auditing can be easily done.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2418", "sentence1": "It may be considered that a camera indicator light may be shown when taking pictures.", "sentence2": "all the experiment phones used by our labmates (see Table 3) do not show it.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2419", "sentence1": "They mainly focus on understanding security risks in vendors' customization process; and their attack is inspired by a security flaw that vendors have mistakenly configured the camera device node with permission of 666.", "sentence2": "at the beginning, our focus is to construct a transplantation attack to evade API auditing.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2420", "sentence1": "To avoid the second defence, the attacker could use a standard picture-taking app instead of QR-code scanner app to add spy-on-user code.", "sentence2": "as the malicious app is already set as a member of the camera group.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2421", "sentence1": "At last, attackers can submit the repackaged app(s) to, e.g., third party markets to dis­tribute it.", "sentence2": "privacy-harming pic­tures can be taken by the malware without the Android API auditing being aware of it.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2422", "sentence1": "For the convenience of test, we let the malicious app show the preview window on the screen.", "sentence2": "to check whether a phone's screenshots (generated by the test platform) contain the preview window or not, we can verify whether the picture taking is successful or not.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2423", "sentence1": "Second, the image data may exist in the kernel space since the camera driver runs in the Linux kernel, and native code running in the user space cannot get the data.", "sentence2": "there are several message types enabled in the mediaserver process when taking a picture.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2424", "sentence1": "The difference between the two is that the first one compares all signatures with each page of memory, whereas the second one scans memory byte-by-byte, reading 32-bits at a time, and identifies more relevant signatures that can be matched with the page.", "sentence2": "we would expect the second approach to be more efficient for working with a large number as it shifts the primary dependence of processing time from the number of signatures to the size of pages.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2425", "sentence1": "Unlike prior work, which relies heavily on deep manual analysis and results in fragile methods that are not guaran­teed to work on newer versions, CodeIdenti.er presents a fully automated solution that is fast, accurate, and robust.", "sentence2": "second, it identifies the kernel code the PGD is used to make clusters of the similar read-only pages, assum­ing that the kernel-code pages are read-only for code pro­tection.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2426", "sentence1": "Conceptually, the filtering does not alter the asymtpotic complexity–in the worst case, all signatures could end up in the same set.", "sentence2": "under the assumption of relatively uniform distribution of the O-P keys (which we have observed empirically), it does reduce the number of signature comparisons by a substantial constant factor. ", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2427", "sentence1": "Since data structure definitions vary across operating systems, Siggraph can be used for OS .ngerprinting.", "sentence2": "the only reliable approach is to directly examine the physical memory of the VM.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2428", "sentence1": "This special case can be detected by simply keeping a crypto hash of the page, and depending on use case turn hash-based page matches on/off.", "sentence2": "the baseline version starts much lower but rises at a much steeper angle than the second algorithm, which filters the candidate signatures before comparison; the crossover point is around 1,000 signatures.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2429", "sentence1": "Several approaches are proposed to automate the triggering of UI events, from random event generation [32] to more advanced approaches like AppsPlayground [39] and SmartDroid [48].", "sentence2": " all of them still have many limitations on the type of events they can handle and the coverage.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2430", "sentence1": "StaDynA extends the initial MCG generated with a tra­ditional static analyzer with the information detected at run­time.", "sentence2": "if an application exposes dynamic behavior all mentioned approaches can benefit from the expanded MCG obtained with StaDynA.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2431", "sentence1": "These ellipses show the cases when the MOIs are resolved and corresponding nodes and edges are added to the MCG.", "sentence2": "the client side can run either on a real device or on an emulator.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2432", "sentence1": "One may attempt to distribute the dataset into multiple hashtables across several machines and coordinate the nodes to compute set intersections for leak scanning.", "sentence2": "such a system is nontrivial to im­plement from scratch and has not been reported in the lit­erature.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2433", "sentence1": "One collection consists of shingles obtained from the con­tent segment and the other collection consists of shingles from the sensitive sequence.", "sentence2": "to compute the intersection rate of two fingerprint collec­tions Irate, we design two MapReduce algorithms, Divider and Reassembler, each of which has a map and a reduce op­eration.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2434", "sentence1": "If the sets are relatively small, then a faster implementation is to use a hashtable to store set A and then testing whether items in B exist in the hashtable or not, giving O(n + m) complexity.", "sentence2": "section 4 presents the security analysis of our approach especially on the con.dentiality of sensitive data.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2435", "sentence1": "We group the nodes that are closest (in terms of number of hops) to one another into an anonymity set.", "sentence2": "given an anonymity set S, we have an ordering of them d1, . . . , d|S| such that di and di+1 are one hop from one another", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2436", "sentence1": "For example, consider an alternative to source-routing, which is table-based routing, but with shortest-paths.", "sentence2": "the client does not provide a route; rather, the nodes in the wireless sensor network are capable of routing by themselves.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2437", "sentence1": "We do not constrain the attacker A in any way other than to require that it is an algorithm.", "sentence2": "the attacker may have access to any information he wants before he starts the game (see, however, the caveat in the next paragraph).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2438", "sentence1": "In MakeLeftHeavy, we reorder the children of every node so that they are ordered left to right in non-increasing height.", "sentence2": "the leftmost child has the maximum height amongst all the node's children, the next child to the right has height at most the height of its left sibling and so on.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2439", "sentence1": "Presumably, the client has placed that node in the same partition as the two nodes colored blue.", "sentence2": "the client queries all three nodes when it intends to query the node shaded black only.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2440", "sentence1": "Our particular intent with the empirical evaluation is to validate the as­sertions on sufficiency in Theorems 4 5, 7 and 8.", "sentence2": "we sim­ulate the DAS protocol, measure the communication-cost it incurs in various settings, and observe that it is indeed upper-bounded as the analytical assertions tell us.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2441", "sentence1": "Clustering is executed on the set of flows that will be used to build the training dataset.", "sentence2": "after performing the clustering the training dataset will be composed as follow.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2442", "sentence1": "In this section we describe our framework.", "sentence2": "section 3.1 introduces the pre-processing steps that allow us to model the network traffic.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2443", "sentence1": "In the last years, several concerns have been raised about the capabilities of those portable devices to invade the privacy of the users and to become actual tracking devices .", "sentence2": "the median value and the third quartile have been used as thresholds to limit the maximum length of the generated time series.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2444", "sentence1": "However, often times intrusions on the database level are not properly reflected in the network or OS level [11].", "sentence2": "a few intrusion detection systems have been proposed that uses the database audit trails (i.e., SQL query trace) to identify an intrusion against the underlying DBMS.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2445", "sentence1": "We generate our queries in such a way that any query on T P is equally likely.", "sentence2": "our query generation procedure is unbiased.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2446", "sentence1": "However, these queries would be easily detected if they are evaluated with respect to the context they are submitted to the database.", "sentence2": "they would be detected by a DIDS which takes into account the previously submitted queries.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2447", "sentence1": "Role threshold values greatly depend on the query behavior of the users of the corresponding role.", "sentence2": "we propose the following heuristic-based threshold generation in this paper.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2448", "sentence1": "A valid user, on the other hand, have incentives to access both sensitive and insensitive parts of the database.", "sentence2": "a database intrusion detection system, when equipped with the knowledge of data sensitivity values, is more likely to distinguish a sequence of anomalous database accesses from sequences of valid accesses.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2449", "sentence1": "However, the query access pattern of an intruder largely depends on data sensitivity.", "sentence2": "an IDS should take into account data sensitivities to isolate intrusive queries from the valid ones.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2450", "sentence1": "This is because, it is usual for Jane to submit queries to retrieve cus­tomer account information.", "sentence2": "these queries would be easily detected if they are evaluated with respect to the context they are submitted to the database.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2451", "sentence1": "Android applications are formed by logically separated components that mainly communicate with each other through Intents, which carry data and request the execution of a procedure to another application.", "sentence2": "the Android Intent Passing mechanisms do not provide the receiving component with any information concerning the origin of an intent, thus facilitating the creation of spoofed intents with malicious input data [1, 2].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2452", "sentence1": "One of them is a method by Sawamura et al. for personal identification in which individuals are identified by four types of verification with a multi-touch panel [8].", "sentence2": "the method has a problem that paths by fingers can be traceable by fingerprints which remain on the special panel since the panel is never touched for purposes except personal identification.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2453", "sentence1": "When the finger horizontally moves plus one millimeter at the first trial, and it horizontally moves minus one millimeter at the second trial, the average is zero although the standard deviation is not zero.", "sentence2": "we eliminate the value of the characteristic from CVK when the value of the average of the characteristic is zero.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2454", "sentence1": "In this chapter, existing researches in which characteristics of personal actions are used for personal identification are introduced.", "sentence2": "m is narrowed, if characters for a password consist of the past candidate characters.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2455", "sentence1": "Some of the characters for the password is presumable when attackers did not gaze at the screen since the arrangement of software keys is fixed and passwords are usually short for convenience.", "sentence2": "we proposed a method for enhancement of security of password lock for smart phones by applying the comparison of personal characteristics when passwords are inputted with flicks.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2456", "sentence1": "However, they choose ten most effective gestures for their method.", "sentence2": "two hundred trials are much enough to break their personal identification method since twenty trials for one gesture is required and there are ten gestures.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2457", "sentence1": "Candidate characters for password decrease according to the increase of the number of registered users.", "sentence2": "we renew the method by using double stage C.V. filtering in order to solve the problem [4].", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2458", "sentence1": "The number of recommended characters for password is shown in Table 2.", "sentence2": "fSi-1 is almost equal to FSi and GVi-1 is equal to GVi.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2459", "sentence1": "There already exists open source software that allows for multiple keys per access to one file.", "sentence2": "both user A s key and user B s key can decrypt their corresponding symmetric keys with the same result.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2460", "sentence1": "The first party would be the DM, preferably an always-on desktop computer secured with the enterprise; the second party would be the DC, a mobile device that would need access to cloud-based data from any location; and the third party would be a third-party Keying as a Service Provider (KaaSP).", "sentence2": "our proposed system does allow for a threshold of at least 2 parties with minimal overhead.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2461", "sentence1": "Larger keys, however, will not solve all the problems because as more devices can handle sufficiently large keys, and more employees migrate their work to mobile devices such as smart phones, the total number of keys that an organization has to manage may continue to increase.", "sentence2": "even though more devices will be able to use longer key lengths, the total number of keys may also become unmanageable for businesses as each employee's device is given its own unique key.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2462", "sentence1": "As mentioned, portable devices are becoming more popular, and computing power is increasing.", "sentence2": "we suggest that access of the files is within a bigger framework that allows for easy and instantaneous return to the work environment upon reauthentication.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2463", "sentence1": "Because, it is for a Vary header returning the contents which change with kinds of User-Agent.", "sentence2": "the public key for creating the Bloom filter which stores a user's ontological kind and numerical value is required N. ", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2464", "sentence1": "Moreover, the number of crimes on Internet increases according as Internet spreads.", "sentence2": "a system for identification of users is required by specific administrators.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2465", "sentence1": "That is to say, users must be identified or grouped by specific administrators while their anonymity must be assured for the third party.", "sentence2": "we propose a method in which information of users' favorites is encrypted and stored in HTTP request header using a Bloom filter.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2466", "sentence1": "On the other hand, the pursuit measure with cookie can discern client computers without depending on IP address.", "sentence2": "website operators, Internet advertisement distribution contractors, etc. use the measure for acquiring detailed access histories of users.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2467", "sentence1": "We expect the ASR performance to be higher when adapting the recognition dictionary to the exact vocabulary (including nonsense speech) of the CAPTCHA.", "sentence2": "this would increase the attacking costs considerably as it requires a large number of labeled CAPTCHAs.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2468", "sentence1": "To be able to perform the required number of observation at­tacks, we randomized pattern generation and simulated user input.", "sentence2": "some real-world factors were not consid­ered.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2469", "sentence1": "In addition, our perception analysis indicates that more complex patterns are perceived hard to enter.", "sentence2": "we assume that \"special moves\" like overlaps and knight moves are hardly used in the wild.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2470", "sentence1": "If the time window is enlarged to 15 minutes, then 99.8% of HTTP requests and their dependencies are included.", "sentence2": "the HTTP requests with dependents are temporally close to each other, despite some rare cases.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2471", "sentence1": "Compared with the original view, our condensed view significantly reduces the redundancy of displaying leaf nodes.", "sentence2": "it helps users identify abnormal nodes due to its visually salient.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2472", "sentence1": "Their main drawback is that, given the enormous number of different traffic types of interest, building a well-defined profile encompassing all normal traffic is extremely difficult.", "sentence2": "they suffer from high false alarm rates.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2473", "sentence1": "Since UBM is trained on a large pool of data, one expects that a higher number of components could lead to a more precise model.", "sentence2": "when the number of components rises tight, we may have no access to enough training data for some of the components.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2474", "sentence1": "In the present snippets, we have inserted an attacker action between the two relevant defender operations.", "sentence2": "we have also identified several small changes to the POSIX API that would make it easier to write race-free code.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2475", "sentence1": "In this case, the PATHR E S call would fail to open the file, also indicating a race.", "sentence2": "no matter what the attacker does, he will be detected.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2476", "sentence1": "This anal­ysis indicates that the FileObserver based light monitoring malScan lacks comprehensiveness.", "sentence2": "all 30 tested AVDs have designed the heavy sweeping malScan operation, which is a comprehensive malware scan.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2477", "sentence1": "Bearing this assumption in mind, we would like to understand if there are still deficiencies in the malware scan (malScan) mecha­nism itself that can cause potential hazards.", "sentence2": "our research sheds the light on the impor­tance of taking the secure and preventive design strategies for AVD or other mission critical apps for fast-evolving mobile-systems.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2478", "sentence1": "In this way, Fsetup always knows from which participant a message comes in, and to which participant a message is to be sent to.", "sentence2": "for participants and their algorithms, there is no guarantee that the “plugging” and/or the input-manipulation runs correctly. ", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2479", "sentence1": "In order to ensure that the VOIP services are always available, the iOS system will relaunch the VOIP app in the background immediately after system boot. ", "sentence2": "unless the customer terminates the app manually, the app can always monitor phone call events even after rebooting the device.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2480", "sentence1": "Note that because there is no app review process for enpublic apps, the app can use any dynamic loading techniques (e.g., NS-Bundle and dlopen()) to load the private frameworks at runtime.", "sentence2": "in addition to checking the static links of frameworks, iAnalytics also analyzes the dynamic loading behaviors at runtime.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2481", "sentence1": "Although IOKit framework does not belong to the private frameworks, there is no public API calls in this framework.", "sentence2": "any API calls in this framework also belong to private APIs.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2482", "sentence1": "Although ad hoc distribution has a 100 device install limit, hackers can rename the Bundle name of the app, iTunes will then treat the renamed app as a new app with another 100 device limitation.", "sentence2": "hackers can use this method to attract customers to download free pirated apps which are not free in Apple Store from their third–party markets, and then use advertisement to gain money. ", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2483", "sentence1": "Cloud application marketplaces of modern cloud infrastructures offer a new software deployment model, integrated with the cloud environment in its configuration and policies.", "sentence2": "similar to traditional software distribution which has been suffering from software piracy and reverse engineering, cloud marketplaces face the same challenges that can deter the success of the evolving ecosystem of cloud software.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2484", "sentence1": "The selected disk image is then written to the virtual disk of the VM, so the application can be used by the cloud users.", "sentence2": "similar to traditional software distribution which has been suffering from software piracy and reverse engineering, cloud marketplaces face the same challenges that can deter the success of the evolving ecosystem of cloud software.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2485", "sentence1": "Our binary rewriter enforces the same set of rules as Native Client (NaCl) [38, 48].", "sentence2": "instruc­ tions are grouped into equal-sized bundles (16 bytes), and indirect branch instructions such as indirect call and return must target the boundary of a bundle.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2486", "sentence1": "It then retrieves the bytecode of the method to decode and execute it.", "sentence2": "dex sandbox is implemented in the Java and C++ programming language, while native sandbox is implemented in C and the ARM assembly.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2487", "sentence1": "specifically, bytecode rewriting-based approaches insert inline reference monitors to reg­ulate apps behaviors.", "sentence2": "dynamically loaded classes (e.g., classes loaded by DexClassLoader) could pose serious challenges to these systems since the dynamically loaded bytecode is not avail­able when statically rewriting the app.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2488", "sentence1": "This trend is accompanied by the vigorous increase of third-party Android apps that are available for users to download and install.", "sentence2": "recent studies reveal that third-party apps, including popular ones, could leak private information without user consent or awareness [12,19].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2489", "sentence1": "It costs about 3.9 seconds in total for this app to complete.", "sentence2": "the average time for each policy retrieval is about 0.39 millisecond.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2490", "sentence1": "Based on the re­cent surveys [5, 14], the total number of IP blocks in a full circuit is around 80 on average.", "sentence2": "we partition each benchmark circuit into 80 blocks, where the size of each block is slightly different.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2491", "sentence1": "And the cost of each simulation is in proportional to the size of circuits, i.e., the number of gates and interconnects [25].", "sentence2": "on a modern PC with an Intel 3.6GHz CPU, it takes nearly an hour on a circuit with about 25, 000 logic gates to perform 64, 000 Monte Carlo simulations [7].", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2492", "sentence1": "Usually, we need two separate proofs that prove the security of SBE scheme under two different modes, Select-mode and Cut-mode, respectively.", "sentence2": "we found that these two proofs has too many similarities according to the comparison result in Table 1.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2493", "sentence1": "Without doubt, the data mining-as-a-service benefits business intelligence.", "sentence2": "there exists a serious privacy issue in the paradigms, that is, the server has access to the data of the company and may learn business secrets from it.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2494", "sentence1": "Consequently, it is only possible to have a fast and complete addition law when a = -1 is a square in the underlying prime field Fp, which is the case if and only if p ≡ 1 mod 4.", "sentence2": "a MoTE curve can only be generated over a prime field Fp whose order is congruent to 1 modulo 4 [6, 16].", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2495", "sentence1": "In 2004, Gura et al [7] published a now-classical paper in which they demonstrated that, in contrast to conventional wisdom, strong PKC is feasible for small battery-powered sensor nodes.", "sentence2": "they showed that Elliptic Curve Cryptography (ECC) [8], when carefully implemented and optimized, is computationally less costly than was believed at that time.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2496", "sentence1": "A serious attacker would never attempt to break the ECDLP in a 160-bit group as he can get the secret key in a much cheaper way via reverse engineering.", "sentence2": "we decided to use elliptic curves with orders of about 2160 and 2192 ; both orders are well established and supported in the WSN research community.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2497", "sentence1": "All previous implementations of ECC for MSP430 processors used either a conventional Weierstraß curve (e.g. [20, 23, 27]) or a Montgomery curve of unreasonably large order (e.g. [10]), both of which wastes execution time and, thus, energy.", "sentence2": "our software supports Montgomery and Edwards curves over a 159 and a 191-bit prime field, which represents a good compromise between performance and security.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2498", "sentence1": "SafeCurves requires f to be at least 2200, in which case computing the ECDLP is infeasible with today's technology.", "sentence2": "a so-called pseudo-Mersenne prime has the form p = 2k - c where c is small in relation to 2k; typically, c is chosen to fit into a register of the target processor.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2499", "sentence1": "Public key based protocols [1, 2, 10] enables the verifier to verify arbitrary anonymous groups without overwhelming computational cost.", "sentence2": "low-cost pas­sive tags usually cannot afford these protocols due to the extensive computation requirements and therefore, we do not consider this type of solutions as it is rather expensive and impractical.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2500", "sentence1": "Second, the use of SYN proxy limits the num­ber of connections that can be forwarded.", "sentence2": "the maximum number of connections forwarded to a specific {I P, port} pair is 64513 (see Section 4.2).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2501", "sentence1": "Indeed, as discussed in Section 4, while being an effective mechanism to protect against SYN flooding attacks, proxying introduces several problems which derive from breaking the end-to-end paradigm.", "sentence2": "its use should be limited as much as possible.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2502", "sentence1": "Note that, according to TCP protocol specifications [5, 13] each ISN is computed in a non-predictable way.", "sentence2": "it is impossible for R to predict the ISN that B would generate and, as a consequence ISNR will be different from the sequence number B will use.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2503", "sentence1": "Breaking TCP end-to-end semantics introduces the need to store state, which in turn opens the system to attacks exploiting buffer saturation.", "sentence2": "there is a strong need to reduce its use as much as possible while retaining its beneficial effects against control plane saturation attacks.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2504", "sentence1": "This number can be quickly reached if we consider extremely popular HTTP services (e.g., Google or Facebook).", "sentence2": "for completeness, we compared the average overhead introduced by Avant-Guard with the overhead introduced by LineSwitch, evaluating both a scenario without attack, and a scenario under SYN flooding based control plane saturation attack.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2505", "sentence1": "For instance, our observation indicates that read and write are among the most frequently used system calls.", "sentence2": "a process must first open a file before it could perform read/write opera­tions on the file, and the usage of open and close system calls are over an order of magnitude less.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2506", "sentence1": "A popular way to find the period from data sequences is via Fast Fourier Transform (FFT).", "sentence2": "the IT group is responsible for managing computers in the classrooms, faculty offices, and student laboratories.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2507", "sentence1": "E-Autocorrelation shares the same property with the conventional autocorrelation, that is, when a periodic event sequence is shifted with a lag that is exactly a multiple of the period, re achieves its maximum.", "sentence2": "e-Autocorrelation is suitable for periodicity detection in event sequences.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2508", "sentence1": "Note that in a perfectly periodic sequence, E-Autocorrelation values as 1 indicate periodicity.", "sentence2": "since the event sequences could be noisy, choosing smaller values than 1 as peaks provides a chance for our approach to better tolerate noises and thus remain robust.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2509", "sentence1": "For category I processes, because of their complete idleness, the constraining will be in effect immediately.", "sentence2": "any system call from these process will be considered as anomaly and thus will trigger manual intervention.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2510", "sentence1": "In this work, we simply choose to design a simple model based on system call instrumentation only to illustrate that idling processes can be easily constrained.", "sentence2": "we intercept all system calls of the target idling process, and build its “idling behavior” profile, which comprises of the variety of system calls as well as their calling context (CC) [36].", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2511", "sentence1": "As shown in Figure 2, The Delivery Mode (DM) of the I/O redirection table can be configured as SMI instead of Fixed normally.", "sentence2": "we are able to deliver an SMI to the CPU for every keyboard interrupt; that is, every key press causes our code to execute in SMM.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2512", "sentence1": "Note that the transmit queue may become empty before finding the login packet, so we may miss the first transmit descriptor that arrives at the empty transmit queue.", "sentence2": "we insert a transmit descriptor whenever the transmit queue becomes empty until identifying the login packet.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2513", "sentence1": "Then, we enable TrustLogin and login to the FTP server again.", "sentence2": "the password recorded by the keylogger has been changed to a random string generated by TrustLogin. ", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2514", "sentence1": "As stated in the paper, higher frequency of polling may affect the proper display of the graphics, and the user may notice the abnormal event.", "sentence2": "trustLogin is able to defend a GPU-based keylogger as long as the polling interval does not pass below 15 microseconds.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2515", "sentence1": "Physical layer message integrity protection and authentica­tion by countering signal-cancellation has been shown as a promising alternative to traditional pure cryptographic mes­sage authentication protocols, due to the non-necessity of neither pre-shared secrets nor secure channels.", "sentence2": "the security of such an approach remained an open problem due to the lack of systematic security modeling and quantitative analysis.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2516", "sentence1": "Through correlated jamming, the attacker has the potential to modify/cancel any signal in wireless channel, and the message integrity will not be protected.", "sentence2": "it is essential to investigate the possibility of signal cancellation in the real-world, so as to provide quantitative security guarantees.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2517", "sentence1": "Observe that in Game 1, all oracle outputs are not related to r * .", "sentence2": "the encryption oracle does not help the attacker to win the game.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2518", "sentence1": "This paper presents a novel deanonymization attack on Tor that exploits a fundamental weaknesses of low-latency anonymization networks.", "sentence2": "we show that an attacker capable of providing web content to users, e.g., through banner advertisements or cross-site scripting, is able to deanonymize users via a sidechannel attack.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2519", "sentence1": "Several au­thors demonstrated the effectiveness of such attacks [6, 9].", "sentence2": "the secu­rity of Tor is based on the use of strong encryption and the large number of relays that can be used to establish a path, thus significantly lowering the ability of an attacker to easily eavesdrop a communication or link senders and receivers.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2520", "sentence1": "Most closely to our approach is Aurasium [39], as it deploys a similar redirection technique: it overwrites entries of the global offset table (GOT) – that holds runtime addresses of lazy and non-lazy symbols – with the start address of policy check functions.", "sentence2": "in contrast to XiOS, it does not provide any mechanism to hide the actual runtime addresses in a shadow table.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2521", "sentence1": "For this, we support different policy enforcement options: allow, deny, log or modify.", "sentence2": "the modify option allows the replacement of the arguments passed to an external function and public API return values.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2522", "sentence1": "R2: Require no changes to the operating system and cur­rent software stack architecture of iOS.", "sentence2": "we need to ensure that private APIs are still accessible from public frameworks.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2523", "sentence1": "As described in Section 4.3, we intercept only calls to external functions.", "sentence2": "the main application code and instructions from shared libraries execute with na­tive performance.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2524", "sentence1": "Although PiOS revealed that many apps leak the device ID to application developers, it cannot detect Jekyll-like attacks [35] where the malicious behavior is only triggered at runtime.", "sentence2": "moCFI [15] and PSiOS [37] could potentially prevent the mentioned attacks by enforcing control-flow integrity (CFI) and fine-grained sandboxing policies.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2525", "sentence1": "To test XiOS against previous attacks, we let our malicious application de-reference the runtime address of a public API.", "sentence2": "the modify option allows the replacement of the arguments passed to an external function and public API return values.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2526", "sentence1": "Apple maintains an app store, first introduced in July 2008, that hosts in Sep. 2014 more than 1, 300, 000 applications (apps) [28].", "sentence2": "the popularity, the high number of features and apps, as well as the large amount of sensitive and private information that are available on iOS devices make them attractive targets for attackers.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2527", "sentence1": "The computation outsourcing techniques [36] enable mobile devices to outsource large computation tasks in decrypting complicated ciphertexts.", "sentence2": "these techniques require the mobile users to interact with a computing server for help to complete the computation tasks whenever the mobile users need to decrypt a ciphertext, which is inefficient due to the frequent interactions with the computing server. ", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2528", "sentence1": "Their scheme mainly focuses on the reduction of decryption com­putation and does not consider the access rights delegation from one to another.", "sentence2": "the costs for conversion key generation and file conversion are also less for more powerful computing devices.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2529", "sentence1": "We can calculate the estimated probability of the event as the fraction of measurements containing this event within the whole observation history database HX for context X. I{Ci .", "sentence2": "due to this same property, the prover and verifier need to be located relatively close to each other in order for their approach to work, lim­iting its practical applicability.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2530", "sentence1": "For example, a device might want to reveal its presence in a particular location only to those peer devices that are present in the same location [10].", "sentence2": "the devices need to be able to verify that a location claim made by a peer device indicating proximity is indeed genuine.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2531", "sentence1": "We only know that this occurrence frequency is smaller than f. ", "sentence2": "we take f as the upper bound for the occurrence probability of the measurement.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2532", "sentence1": "It is difficult for external entities to determine whether the location claimed by a client device is in fact correct.", "sentence2": "there is a need for location proofs: methods for verifying the correctness of location claims that clients present to the LBS. ", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2533", "sentence1": "For example, if an application registers as a broadcast receiver with the NEW_OUTGOING_CALL action, when an outgoing call event occurs, the Android system broadcasts the event, which notifies the event to the receivers registered with the NEW_OUTGOING_CALL action.", "sentence2": "our malware can catch call actions without registering as a receiver.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2534", "sentence1": "Unlike static analysis which requires precise string values that particular variables may have, taint analysis simply tracks value flows be­tween sources and sinks.", "sentence2": "the taint analysis performs only a flow-insensitive analysis which provides sound analysis results.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2535", "sentence1": "Note that our fake malware does not send private user data to our server; it merely sends bogus values to our server.", "sentence2": "it did not leak any private user data from anyone happened to download it.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2536", "sentence1": "Because ADB is originally for debugging purposes, the Android system assigns higher privileges to ADB than to third-party Android applications.", "sentence2": "we could present powerful attacks of various kinds by leveraging ADB and its utility functions only with the INTERNET permission.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2537", "sentence1": "Effort has been taken to ensure that the core security strategies and underlying issues un­covered remain intact.", "sentence2": "details such as IP addresses are anonymised.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2538", "sentence1": "In contrast, the pri­vate charging attack mainly utilizes a private charger, which focuses on a smaller ranger of targets.", "sentence2": "the private charger may obtain a higher credence from people than the public chargers, so it is possible to discover more sensitive information through social networking (i.e., asking people to show something).", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2539", "sentence1": "In this section, we thus develop juice filming attacks, which are a type of charging attacks but without the need to request for phone unlock and any permissions.", "sentence2": "whether users trust or distrust the computer does not affect our attacks.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2540", "sentence1": "In Figure 4 (a), we present a full and explicit implementation of our attack by means of VGA2USB.", "sentence2": "another requirement of our attack is that users have to interact with their phones during the charging, so that our attack can record all inputs.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2541", "sentence1": "The paper includes an appendix that introduces two additional protocols.", "sentence2": "the BC sets the session number i, updates nodes secret keys, and creates a session key SK with header hdr.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2542", "sentence1": "The problem of key establishment and agreement for hardware-limited networks is widely discussed in the literature, and Kim et al. [10] proposed a systematization of security properties for these protocols.", "sentence2": "a majority of the proposed methods provide security properties in the face of weaker adversaries than these considered in our work.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2543", "sentence1": "For our purpose multi-cloud is a cloud federation where all collaborating members are exclusively clouds.", "sentence2": "we do not consider a cloud and an external identity provider to constitute a multi-cloud.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2544", "sentence1": "Backes et al. [5] and Gionta et al. [25] later proposed two execute-only memory solutions: XnR and HideM respectively.", "sentence2": "these two approaches do not consider the threat of pointer harvesting.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2545", "sentence1": "That is, a function return is only allowed to return to its original caller.", "sentence2": "several solutions use heuristics to compensate for the coarse­grained protection of returns, e.g., by monitoring the num­ber of instructions executed between a pre-defined number of consecutive indirect branches [40, 13].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2546", "sentence1": "Davi et al. [18] conduct a systematic security analysis of the recently proposed CFI solutions including kBouncer [40], ROPecker [13], CFI for COTS binaries [58], ROPGuard [23], and Microsofts’ EMET tool [36].", "sentence2": "they derive a combined CFI policy that, for each type of indirect indirect branch and behavioral heuristic (e.g., the number of instruction executed between two indirect branches), uses the most restrictive setting among the aforementioned policies.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2547", "sentence1": "The literature on control-flow integrity is substantial.", "sentence2": "we limit our discussion to recent and unconventional approaches.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2548", "sentence1": "The assumption underlying all these works is that the disclosure of a single address no longer allows an adversary to launch a code reuse attack.", "sentence2": "as will be explained in Section 2.1, more involved types of memory disclosure vulnerabilities can be exploited to bypass fine-grained code randomization.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2549", "sentence1": "As with many real-world ROP attacks, the disclosure of a single runtime memory address is sufficient.", "sentence2": "in contrast to standard ROP attacks, JIT-ROP does not require the precise knowledge of the code part or function the memory address points to.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2550", "sentence1": "If such sensitive information is leaked out of the browser without protection, it can be used by unauthorized parties to illegally access users’ online accounts, steal their online identities, or track their online behaviors. ", "sentence2": "banning extensions that may leak users’ sensitive information is also necessary and important.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2551", "sentence1": "If such sensitive information is leaked out of the browser without protection, it can be used by unauthorized parties to illegally access users' online accounts, steal their online identities, or track their online behaviors.", "sentence2": "because the execution traces only contain the call relations and do not contain any information from users, they can also be shared (e.g., in a repository along with the extensions)among the analysts to further cover more execution paths of the extension.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2552", "sentence1": "Thus even if the user Ui enters his/her wrong password and/or wrong identification, which may be the identification of another registered user, by mistake, both the login phase and authentication phase are still continued, and finally, at the end of authentication phase, S rejects the Ui login request. ", "sentence2": "it causes unnecessary extra communication and computational overheads during the login and authentication phase.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2553", "sentence1": "However, sharing may cause some undesirable phenomena such as unauthorized access and inconsistent status of shared resources.", "sentence2": "password-based authentication schemes have been widely adopted to protect the resources from unauthorized access.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2554", "sentence1": "The attacker can replay the intercepted messages to fool the remote server Sj.", "sentence2": "in our proposed scheme, we used to generate timestamp that makes synchronization between the user and the remote server.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2555", "sentence1": "In case, such signals are transmitted in analog from, recovery must be possible from the analog form, presumably at a minimum after the signal has been attenuated, distorted and transformed in the process of transmission and reproduction.", "sentence2": "in the case of analog video signals with their high bandwidth requirements, the recovery must then either be possible given only a very limited high fidelity recording of the original signal, or from a significantly lower bandwidth recording at a later stage.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2556", "sentence1": "The image watermarking techniques that are available in the literature are reviewed in brief.", "sentence2": "in the case of analog video signals with their high bandwidth requirements, the recovery must then either be possible given only a very limited high fidelity recording of the original signal, or from a significantly lower bandwidth recording at a later stage.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2557", "sentence1": "When a node comes up on a network and wants to assign a new address, it must first validate that no other node on the local-link uses this particular address.", "sentence2": "the node should ensure that there is no duplicate of this IP address already live on the network.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2558", "sentence1": "The nature of most identity theft cases are caused due to a shortcoming in the security design, as it is not designed to verify the identity of a user remotely.", "sentence2": "this leads to the compromise of identities causing malicious or fraudulent activities.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2559", "sentence1": "Some behaviorial patterns also .nd application in iden­tity association such as voice modulation and acoustics, the mechanics of locomotion and one s penmanship.", "sentence2": "the nature of a per­son s walk is contingent to various situations and scenarios, and it is extremely difficult to generate consistent results over a person s lifetime.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2560", "sentence1": "The proposed PHP-Sensor defensive model for discovering workflow violation attacks has two main phases: offline and recognition phase.", "sentence2": "for the detection of XSS worms on the client-side user interface, PHP-Sensor operates only in recognition phase.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2561", "sentence1": "Grouping: Usually two web pages are possibly produced from the similar web page pattern if they consist of related set of possible crucial paths.", "sentence2": "the malware of an XSS payload may possibly be transmitted in the shape of text message or in the form of a URI web link directing to the outward file kept on a remote location of server.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2562", "sentence1": "In the later phase, the deduced model is utilized to assess every incoming HTTP request and outgoing HTTP response and discover any variations.", "sentence2": "for the detection of XSS vulnerabilities in PHP web applications, our defensive model circumvents the dissemination of XSS worms by observing the outgoing HTTP web request that transmit self-propagating payloads.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2563", "sentence1": "Mobile and embedded system software designer are often torn between choosing security and functionality.", "sentence2": "the se­curity of out-of-band execution environment is sensitive to rich functionality.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2564", "sentence1": "We report on a series of experiments in prediction for Windows post-release vulnerabilities.", "sentence2": "vPMs operate on highly unbalanced datasets.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2565", "sentence1": "By reading the input validator code, the players can easily infer the input validator type (e.g., credit card number, email address, phone number, SSN, URL, and zip code) even if the type is not explicitly stated in the playervisible code segment.", "sentence2": "the players may not immediately realize the exact expected behaviors of the given input validator for that type.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2566", "sentence1": "Stakeholders of such systems demand maximum reliability and security.", "sentence2": "the methodology must estimate the security of a system in terms of the loss that each stakeholder stands to sustain as a result of security breakdowns, the so-called mean failure cost (MFC) [8­10].", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2567", "sentence1": "Next, we show how to perform homomorphic operations on this HTDF.", "sentence2": "we view HTDFs as providing an interesting conceptual uni­fication of homomorphic signatures and encryption.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2568", "sentence1": "One advantage of the SNARK-based scheme is that a signature can be verified very efficiently, independently of the complexity of the computation f being verified, as long as f has a short Turing-Machine description.", "sentence2": "in this work, we will only get efficient verifi­cation in an amortized sense, when verifying a compu­tation f over many different datasets.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2569", "sentence1": "Second, for concrete­ness we chose our construction to be in the Turing Machine model, however, we believe it could be fairly easily adjusted to a different model such as RAM by simply letting the next position be a function of the state into [0, S-1] as opposed to moving the head position.", "sentence2": "our proof struc­ture would remain the same with this minor change to the construction.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2570", "sentence1": " In the TLS PKI, a certificate authority (CA) signs a public-key certificate that binds a server name to a public key, and a client can verify this certificate during TLS connection setup.", "sentence2": " As with CAs, vendors can change security indicators for specific non-deploying domains.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2571", "sentence1": "A vendor can add browser features that effectively deploy a proposal at all of its customers, and few vendors need to deploy a proposal to spur adoption among a majority of Internet users.", "sentence2": "vendors should exert their strong influences over other parties to most effectively deploy a proposed PKI enhancement.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2572", "sentence1": "Domains that deploy a PKI proposal can incentivize clients to deploy the new scheme by offering a more secure service, particularly when the domain stores private user data.", "sentence2": "unlike domains, they generally have no TLS public keys of their own and thus cannot even watch for suspicious certificates for even a single domain without knowing the domain's authentic key a priori.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2573", "sentence1": "In this section we provide background information on log-based PKI proposals for readers.", "sentence2": "we describe the parties in these proposals and provide an overview of the general log-based approach.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2574", "sentence1": "Clients can also influence domains by switching to a different service that does offer the security benefits provided by deploy­ing the new PKI, particularly if the domain provides a security-sensitive service such as e-banking.", "sentence2": "if there is no similar service to switch to, then the clients cannot exert this influence, and thus we claim that clients have a weak influence on domains.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2575", "sentence1": "These documents should be believable and appear to be legitimate.", "sentence2": "adver­saries should have a difficult time discerning decoy files from authentic ones.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2576", "sentence1": "Decoys must also lack any shared attributes that an attacker could test for to discern true documents from fake ones.", "sentence2": "there should be a high degree of variability between decoy documents.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2577", "sentence1": "Since the Panopticlick initiative [8] by EFF, users have become more aware of web tracking and how it affects their privacy.", "sentence2": "there is some pressure on web browser developers to make tracking users more difficult, such as by removing unnecessary sources of entropy in the browser or by deliberately introducing randomness in properties typically used for fingerprinting [17].", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2578", "sentence1": "The first block of declarations (lines 1-11) specifies the attributes of users, resources, and terminals. ", "sentence2": "the user attributes are organization and clearance (line 8): the former contains the name of the organization to which the user belongs, for simplicity, only two values are considered (line 7); the latter is the user clearance level (line 1). ", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2579", "sentence1": "Simple separation-of-duty constraints are user-independent, and it appears most constraints that are useful in practice are user-independent [7].", "sentence2": "cardinality constraints and binding-of-duty constraints are userindependent.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2580", "sentence1": "Similar considerations arise very naturally when we con­sider workflows.", "sentence2": "we may specify authoriza­tion policies and constraints that mean a workflow specification is unsatisfiable.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2581", "sentence1": "In this section, we describe the experimental work on Valued WSP that we have undertaken.", "sentence2": "we will compare the per­formance of our PBB algorithm to that of the state-of-art commercial MIP solver in our computational experiments on Valued WSP.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2582", "sentence1": "In this paper, we describe our ReBAC implementation, discuss the system engineering lessons learnt as a result, and evaluate the experimental work we have undertaken.", "sentence2": "we compare the performance of the various authorization schemes we implemented, thereby demonstrating the feasibility of ReBAC.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2583", "sentence1": "There is now a need for finer-grained access control: e.g., my pa­tient record shall only be accessible by the clinicians who are actually treating me.", "sentence2": "access is granted on the basis of how the requestors are related to me.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2584", "sentence1": "Yet, as we shall see below, ReBAC authorization checks also make use of such relationships when an authorization decision is computed.", "sentence2": "these relationships constitute part of the protection state.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2585", "sentence1": "Of course, this raises the question of why existing PDPs are designed in the way they are.", "sentence2": "we establish a new way of thinking about request evaluation and alternative designs for PDPs in ABAC systems.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2586", "sentence1": "Finally, a request may be malformed and policy evaluation may fail unexpectedly.", "sentence2": "the PDP may return an inconclusive result, indeterminate results, or inconsistent results.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2587", "sentence1": "Let us also point out that the probability does not effectively appear in the request, which means that the previous evaluation functions can still be applied.", "sentence2": "probabilities can always be ignored if they are not relevant.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2588", "sentence1": "Based on these observations, we can conclude that ACDM performs better than existing access control models in that it provides full protection from data leakages with lower efforts on the user side.", "sentence2": "it minimizes the number of policies a user has to write while allowing full coverage of the intended protected set.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2589", "sentence1": "SNOMED-CT also provides a number of attribute relationships which define interrelationships between concepts.", "sentence2": "an attribute relationship is an association between two concepts describing an intrinsic property of the concepts.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2590", "sentence1": "This is mainly due to the significant cognitive burden required by existing languages for pol­icy authoring.", "sentence2": "the access control model is based on a semantic approach which leverages knowledge about the application domain for access decision making.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2591", "sentence1": "In contrast, our approach enables to reason on information inference at any level of the data hierarchy.", "sentence2": "we have tackled the problem of inference control by proposing an access control model that adopts semantic inference relationship among data and defines on top of these relationship authorization propagation rules that prevent inference of sensitive information.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2592", "sentence1": "First, we propose an access control model that prevents inference of sensitive information caused by the semantic relations between data.", "sentence2": "the access control model is based on a semantic approach which leverages knowledge about the application domain for access decision making.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2593", "sentence1": "Second, we demonstrate that the proposed access control model can be implemented using existing access control mechanisms.", "sentence2": "we show that the model can be automatically translated in XACML [33], the de facto standard for the specification and enforcement of access control policies.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2594", "sentence1": "However, these negative authorizations are not correctly interpreted by ACDH2 and ACDH3 .", "sentence2": "they are propagated both up and down the hierarchy in ACDH3 and up the hierarchy in ACDH2 , thus restricting the access to data elements which the user wants to disclose.\n", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2595", "sentence1": "In this section we demonstrate how the proposed access control model can be implemented using existing access control mechanisms.", "sentence2": "we present an encoding of the access control model in eXtensible Access Control Markup Language (XACML) [33], the de facto standard for the specification and enforcement of access control policies.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2596", "sentence1": "specificity can be implemented using the first-applicable combining algorithm.", "sentence2": "the policies associated with a data element and with the data elements that can be inferred from it can be combined using deny-overrides.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2597", "sentence1": "If the original file spans several units, Ul must decode each unit separately in order to read the entire file.", "sentence2": "for each unit, he uses the set of endorsed tokens he can fetch to recover the secret key via CRSS.Combine(·) and then uses the secret key to decode the unit via SFD.Decode(·).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2598", "sentence1": "Even though the cloud promises a convenient way for users to share files and effortlessly engage in collaborations, it still retains the notion of individual file ownership. ", "sentence2": "each file stored in the cloud is owned by a single user, who can unilaterally decide whether to grant or deny any access request to that file.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2599", "sentence1": "3.5 Commune: Protocol specification Recall that Commune leverages a shared repository, which is an abstraction of the owners storage space on S. The shared repository uses a versioning system so that content cannot be overwritten but only new content can be added.", "sentence2": "commune optimizes performance by splitting a file in smaller units, and encoding/decoding each unit sep­arately.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2600", "sentence1": "We note that the security of the un­derlying block-cipher prevents the adversary from recovering partial bits of any cleartext block.", "sentence2": "the adversary can only learn entire blocks of cleartext.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2601", "sentence1": "Second, malicious writers (i.e., writers who have been granted write access by fewer than t owners) are unable to distribute a file without honest readers detecting it.", "sentence2": "a file is considered as written if and only if it is correctly encoded in tokens and those tokens are distributed to and endorsed by at least t out of n owners.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2602", "sentence1": "The latter addresses the case where at a time t1 a user has access to t or more endorsed tokens of a file unit Fi, but at a time t2 > t1, his access rights are revoked.", "sentence2": "at the time t2 , the user has access to fewer than t endorsed tokens.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2603", "sentence1": "As argued before, we cannot prevent users from caching a local copy of the file and reading it at later time when their read rights may have been revoked.", "sentence2": "we still want to provide revocation of a user who only stored the encryption key at the time when he had read access to the file.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2604", "sentence1": "Kerschbaum [21] even used the memory (up to 64k) of tags to store the signatures during the movement of an object.", "sentence2": "it is difficult to store business data in a tag due to the technical limitations and cost issue.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2605", "sentence1": "For example, Wal-Mart itself contains producers, carri­ers, warehouse managers, and wholesalers in a supply chain.", "sentence2": "the proposed framework extended to MA-ABE can be used.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2606", "sentence1": "Especially, the tags must be attached with all merchandise, the cost of each tag is critical for a tracking system.", "sentence2": "it is strongly motivated for a holder of a tracking system to reduce the technical features except for the identi.cation of RFID tags.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2607", "sentence1": "Recent advances in the Internet of Things (IoT for short) have enhanced the efficiency of current tracking systems [6][13].", "sentence2": "these systems are now being widely deployed.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2608", "sentence1": "If a user has two permissions p1 and p2 among other permissions, and uses p1 far more often than p2, one optimization metric is to place p1 and p2 in different roles.", "sentence2": "the optimization metric would be to generate roles that include permissions with similar usage patterns.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2609", "sentence1": "In this experiment, we changed the number of users and permissions, and initial roles used in data set genera­tion.", "sentence2": "none of the proposed methods in the literature allows the administrator to generate a cus­tomized RBAC state based on a customized metric.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2610", "sentence1": "In this example both function calls role_change_permission() in Listing 6.a and reset_role_capabilities() in listing 6.b insert information into the security sensitive table role_capabilities.", "sentence2": "they both have the same SSO.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2611", "sentence1": "Two such groups may require different access control authorities even when they share some common SSOs.", "sentence2": "from the perspective of a given SSO, it has different access control checks when it belongs to different groups.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2612", "sentence1": "In this paper, we want to shine a light on this concern by studying the feasibility of applying ABE on smartphone de­vices.", "sentence2": "we implemented AndrABEn, an ABE library for Android operating system.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2613", "sentence1": "Similarly, the CP-ABE KeyGen operation requires less than 2 s to be executed.", "sentence2": "adopting a security level of 112 or 128 bits, the time overhead imposed by CP-ABE on Android smartphone is much higher, while we argue being still usable for non-interactive applications (e.g., encrypted data to be uploaded to a cloud storage service).", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2614", "sentence1": "Along the same line of studies, we also proposed a system for efficient software updates distribution, over untrusted distribution networks in [5].", "sentence2": "as we can see, the execution time for CP-ABE with AndrABEn is significantly lower compared to the results obtained with our Java based implementation, that in turn presents results that are consistent with the ones provided in [21].", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2615", "sentence1": "They obtained an average decryption and encryption time of 254 ms and 926 ms respectively, considering access structures containing one to five attributes.", "sentence2": "the authors did not provide an implementation of ABE on the Android smartphone.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2616", "sentence1": "How do we know that an email we received from a friend or colleague was actually sent by them?", "sentence2": "most of the early gateways were rather restrictive in their model of access and development.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2617", "sentence1": "Embedding DRAM onto the package adds cost because eDRAM requires additional fab process steps compared with eSRAM.", "sentence2": "when it comes to bandwidth-intense applications, eDRAM has proven to be beneficial for SoC as major area savings offset the process cost.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2618", "sentence1": "As it is shown, the middle plot performed at the selected write duty-cycle, Writeb, has a 50.15% average fractional Hamming weight, indicating that the PUF is perfectly unbiased.", "sentence2": "the top plot performed at Writea has a higher average fractional Hamming weight, due to a longer write duty-cycle which results in fewer write failures.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2619", "sentence1": "There has been an influx of mobile phones and tablets onto today's market.", "sentence2": "it is important for the purposes of the study to spread out the samples collected for neural network training over a large period of time, and this cannot be done when simply taking the first 8 samples.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2620", "sentence1": "Alternative approach is to use public-key cryptography for solving the inherent scalability problem with symmetric-key based schemes, as in [19, 20, 18].", "sentence2": "the tag then sends {C1, C2, C3} to the reader.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2621", "sentence1": "Compared to [11], our work requires 2X the minimum frequency to process videos at a certain frame rate (15.6 MHz vs. 7 MHz, and 124.8 MHz vs. 56 MHz) than [11].", "sentence2": "our architecture achieves 6X gate count reduction compared to [11].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2622", "sentence1": "The present study concentrates on the second view to assess regional development.", "sentence2": "these proposals emphasize the complex nature of regional development by breaking the quality of life in a region in several dimensions.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2623", "sentence1": "c) object field manipulation instructions: these instructions are probably the most common instructions in an app.", "sentence2": "this credential is not used by the app.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2624", "sentence1": "After identifying sink methods, we search for invocations of those methods in the app.", "sentence2": "we locate where those methods are called.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2625", "sentence1": "Our data set has 237 candidate apps in this category, and 121 (51.1%) of them were found to be vulnerable.", "sentence2": "65 of these vulnerable apps were downloaded from Google Play with 11 apps having more than 50, 000 downloads.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2626", "sentence1": "If a mapping still cannot be decided, we preform the manual analysis as the last resort.", "sentence2": "it is much easier to revoke individual credentials.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2627", "sentence1": "Recently, the extensive list of possible Android permissions (≈130 [36]) was restructured to a simpler set of permissions groups, together with a brief summary regarding the different kinds of data an application will have access to [45].", "sentence2": "users must still review the list of permissions every time they install an application or update an application whose permissions have changed.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2628", "sentence1": "There is a clear change in privacy concerns in recent years, as it is easier to share data with others, especially on Facebook [34].", "sentence2": "an open question remains whether users willingness to provide their location and Internet access to applications is in itself a privacy concern.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2629", "sentence1": "Our work has a similar motivation to Kepler [52], a browser extension that monitors and maps network data access, to raise users' awareness on widespread adware and spyware on desktop machines.", "sentence2": "on smartphones, mobile applications are provided with structured APIs that supply a significantly larger amount of sensitive user and sensor data when compared with browsers on desktop platforms.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2630", "sentence1": "Securacy offered access to information previously inaccessible on the mobile phone: a crowdsourced mapping of mobile applications' network connections and their security status.", "sentence2": "malware does exist for iOS, but only 5% of iOS vs 20% of Android users are concerned about their privacy [7].", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2631", "sentence1": "Our participants are somewhat uncomfortable with sharing some data with applications, particularly their contact information (62.3%), profile data (61%) or access to their messages or call data (60%).", "sentence2": "they expressed less discomfort providing access to their browser history (37.6%), Internet connectivity (38.5%) and calendar data (45.8%)(Kruskal-Wallis .2(8)=59.74, p<0.001).", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2632", "sentence1": "Our participants are generally comfortable sharing some data and resources with applications, including Internet connectivity (83.3%), location (80.0%) and profile data (70.0%).", "sentence2": "they are least comfortable sharing access to their messages or calls (6.6%), their user accounts and contact list (23.3%) or their documents (30.0%) (Kruskal-Wallis .2(8)=76.5, p<0.001).", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2633", "sentence1": "Sellwood and Crampton [47] discovered that permissions on newer APIs (including those exclusive to Google developers) would be automatically granted if the application was installed on a device prior to upgrading Android to a newer version.", "sentence2": "a malicious application could tentatively request future permissions (easily retrievable from Android’s open-source code) and wait until the phone is upgrade to support them.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2634", "sentence1": "On Android, users can check the volume of traffic sent but do not receive a breakdown of what and how data was sent and to where.", "sentence2": "it is challenging for user to recognize a suspicious application that may be generating unwanted and potentially harmful traffic.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2635", "sentence1": "Since only exported components are accessible by external apps, these paths cannot be found without performing inter-component communication analysis.", "sentence2": " the statement on line 20 removes d and d.f from the set, hence sink4 on line 23 would not give any warning.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2636", "sentence1": "Some service providers help developers advertise apps on their websites and charge for writing and publishing reviews.", "sentence2": "such paid reviews are forbidden by app store ven­dors like Google and Apple.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2637", "sentence1": "In contrast, for random (i.e.,randomly chosen) apps, none of their reviewers have reviewed more than three apps in the chosen set (we do not show a .gure here).", "sentence2": "active reviewers of promoted apps are likely recruited reviewers.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2638", "sentence1": "Compared to Fig. 11, the average rating distribu­tions of suspicious reviewers are similar in both scenarios.", "sentence2": "the suspicious reviewers reported by our tracer have the similar rating characteristics as recruited reviewers.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2639", "sentence1": "Those reviewers who have reviewed many promoted apps and given high rating scores are highly likely recruited reviewers.", "sentence2": "given a set of recruited reviewers, the apps rated by many of them are also highly likely promoted apps.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2640", "sentence1": "On the other hand, compared to Fig. 6, the average rating distributions of known promoted apps and suspicious apps are nearly identical.", "sentence2": "these suspicious apps have similar rating characteristics as known promoted apps, which indicates highly likely they are also promoted apps.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2641", "sentence1": "Based on these observations, we design an iterative algorithm to trace apps promoted by recruited reviewers.", "sentence2": "one to three months after a new version release have been the common time for web promotion.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2642", "sentence1": "With over a million apps in an app store, it is extremely difficult for a new app to acquire initial reviews, while with no or few reviews an app would be ranked far behind and look much less attractive to app users.", "sentence2": "their re­view promotion services could help app developers booster up their apps out of this difficult time.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2643", "sentence1": "As illustrated in Fig. 4(c), more than 80% apps have less than 10 reviewers in a week and more than 90% apps have less than 50 reviewers.", "sentence2": "most promoted apps have few reviewers in weeks before promotion.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2644", "sentence1": "In our previous comparison between reviewers of promoted apps and reviewers of randomly chosen apps, we observed that no reviewers from randomly chosen apps have reviewed more than three apps from the app set.", "sentence2": "review­ers of promoted apps have reviewed far more apps (up to 246).", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2645", "sentence1": "In contrast, less than 30 random reviewers have reviewed at least 100 apps in the app store.", "sentence2": "overall suspicious reviewers have reviewed significantly more apps than random reviewers.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2646", "sentence1": "Since a large number of packets typically need to be trans­mitted by the attacker before injection can be successful, a spike in network activity might be .agged by an Intrusion Detection System (IDS) as unusual, and the attacker could be blocked consequently.", "sentence2": "in our experience, trans­mitting a large number of frames is not always necessary for the attack to succeed.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2647", "sentence1": "Section 3.6.1 gave an example of how our frame injection attack can be applied if the attacker has access to a service on the internal wireless network.", "sentence2": "in practice this scenario is not very prevalent.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2648", "sentence1": "A similar vulnerability in these kind of aggregated frames would allow an attacker to craft their own MSDUs.", "sentence2": "we determined that this aggregation method is not vulnerable to our injection attack.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2649", "sentence1": "However, the leakage of different bits could be affected by the other, for example cross-talk during the bit transition.", "sentence2": "we would like the observe the result on different subspaces.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2650", "sentence1": "We intro­duce a new SCA profiling method based on Support Vector Regression (SVR).", "sentence2": "sVR aims at building a precise leakage model with given training set.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2651", "sentence1": "We also use Dijkstra from MiBench [8], and parallel versions of Breadth First Search (BFS), Prefix Scan, and Matrix Multiply.", "sentence2": "the SoftBound version has a much higher off-chip access rate be­cause the shadow data structure for bounds metadata resides in the off-chip memory, and must be fetched each time a bound check is performed.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2652", "sentence1": "Bound checking prevents the buffer overflow at­tack.", "sentence2": "it incurs performance overheads associated with managing the security metadata and checks.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2653", "sentence1": "This approach takes advantage of the fact that neural nets only need to be trained once per program, and although they have slow training, they have fast activation.", "sentence2": "although training time is in the order of minutes, prediction can be done in about 210us on our setup.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2654", "sentence1": "Today, the scheduler has limited insights into the opera­tion of the applications that it is scheduling, and detailed software-based analysis of programs may be prohibitively expensive.", "sentence2": "a scheduler has to make best-effort deci­sions about how to schedule applications.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2655", "sentence1": "The attacker is the follower who selects an optimal attack plan based on (limited) knowledge about the defender's strategy.", "sentence2": "we assume that the attacker learns the num­ber and type of deployed HPs; however he does not know which specific hosts are HPs and which are real.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2656", "sentence1": "Secondly, our model can be further extended from the game-theoretical perspective and use additional uncertainty about the knowledge of the attacker, or model multiple types of the attacker using Bayesian variants of Stackelberg games.", "sentence2": "we assume that the attacker learns the number and type of deployed HPs; however he does not know which specific hosts are HPs and which are real.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2657", "sentence1": "Given the intrinsic nature of smartphones and tablet com­puters as mobile devices, their use is almost always linked to the actions of human individuals.", "sentence2": "data stored on or transferred to or from such mobile devices automatically become personally identifiable information, meaning that a direct link can typically be drawn from the data to the in­dividual user of the mobile device.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2658", "sentence1": "As a consequence, such data falls under the privacy and data protection regulations, and have to be processed in compliance with existing data protection laws.", "sentence2": "legal domains are always bound to specific territorial areas (countries, federal states, etc.), which are not always following coherent lines when it comes to data protection regulations.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2659", "sentence1": "Malicious applications that target at the confidential information increase significantly.", "sentence2": "second, make use of the APIs provided by Android SDK for data extraction and communication.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2660", "sentence1": "We will also analyze security activity changes and frequency change between companies after domestic security accidents.", "sentence2": "security activity increased slightly in the industry of manufacturing and construction which have importance in industry of KOREA after domestic security accident but they had an insufficient support and investment of security activity.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2661", "sentence1": "This is likely because the control mechanism triggers more position perceptions regarding the app, including perceived fairness of the app developer, a better general attitude toward the app, more trust of the app, as well as an enhanced intention to use the app in the future.", "sentence2": "giving users actual control over their own information greatly elevated their self-agency, thereby improving their perceptions of the app, which eventually led to more installation.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2662", "sentence1": "In sum, our results show that empowering users with control over information disclosure and informing them about how their information will be used by the mobile app can improve their overall impression of the app and increase their installation behavior.", "sentence2": "such findings are not unidimen­sional; the effects of awareness may depend on whether users are allowed to control their own data flow.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2663", "sentence1": "Many users do not have a good understanding or awareness of how their personal information is shared through mobile apps.", "sentence2": "they are unable to accurately assess the con­sequences associated with their app installation and usage.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2664", "sentence1": "When the num­ber of key apps is within the range from 10 to 49, the aver­age success rate reaches 95% and above.", "sentence2": "when the key apps exceeds 50, the average success rate drops to 90%.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2665", "sentence1": "Issues with side channels: Under the assumptions that stackbase address randomization is not employed and that the key is not changed when restarting a program and Sovarel et al. [38] use timing side-channels to determine the key value on Intel x86 architectures.", "sentence2": "one key factor in their attack was the assumption that the remote machine is running a certain version of an Apache server, whose behavior is used as a side-channel.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2666", "sentence1": "The proposed scheme uses Security Assertion Markup Language (SAML) to provide the Single Sign On functionality.", "sentence2": "the scheme allows the users to register once at the IdP and avail services from another service provider or multiple service providers without registering individually at each Service Provider.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2667", "sentence1": "In the running example, a regular vehicle component may not want to share sensitive owner information with everyone the access should only be granted to police vehicles.", "sentence2": "alternatively, the parameter may be required to be constant, which means that the value must be set at the time of deployment (lines 1, 4) in this case, the certificate for the component states not only its role, but also permitted values of the parameter.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2668", "sentence1": "Moreover, the user does not need to memorize the second secret because it is randomly generated by the CDial, and it is shared with the user at each authentication time.", "sentence2": "the system vibrates briefly when the cursor approaches at a number slot.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2669", "sentence1": "Some of the schemes require an additional device.", "sentence2": "then, we consider a secret input scheme that uses a fixed secret, while an observable input scene appears to be randomized over multiple sessions.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2670", "sentence1": "The input-state indicator represents the PIN input status.", "sentence2": "this scheme increases the input operation load and also requires an additional device.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2671", "sentence1": "One method of realizing the idea is to use a one-time password.", "sentence2": "this scheme increases the input operation load and also requires an additional device.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2672", "sentence1": "During the stage of enumeration, an attacker will try guessing valid credentials for the systems on the network or even discover routing rules and can also collect more information about potential vulnerabilities of the nodes of the network.", "sentence2": "choosing the Tomcat server was no random choice.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2673", "sentence1": "The result from the virustotal service is used as input to the report that is generated and sent to the administrator via email.", "sentence2": "active defenses are promising techniques that are based on a proactive strategy where one anticipates attacks and prepares for the neutralization of the threats.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2674", "sentence1": "Then a linearSVC classifier model is built on these train feature vectors which consist of known malware and benign apps.", "sentence2": "then an app X can be mapped to this space by constructing a vector f(X) , for each feature s extracted from X the corresponding dimension is set to 1 and all other dimensions are set to 0.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2675", "sentence1": "Misuse detector is specifically designed to detect known malware, leading to low number of false alarms.", "sentence2": "misuse detector could not detect zero-day malware.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2676", "sentence1": "Similarly, the methods RiskRanker [7] and DroidAPIMiner [12] use machine learning techniques to detect malware with features statically extracted from Android applications.", "sentence2": "also the analysis results of submitted app are stored in a database in our proposed method.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2677", "sentence1": "(Even when a VM is paused during live checkpointing, the network driver can still buffer incoming packets for that VM.)", "sentence2": "no packet losses happen on VM1 or VM2.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2678", "sentence1": "The comparison between the user's input and the reference template can be performed by adopting standard gesture recognition techniques.", "sentence2": "template-based gesture recognition algorithms, which are currently the dominant ap­proach for interactive gesture recognition applications, rely on similarity measures which can be used to compare the two different gesture inputs.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2679", "sentence1": "In our case authentication is unidirectional, i.e., the device that wants to prove proximity is the one displaying the authentication gesture whereas the device verifying proximity is the one capturing the user gesture and internally comparing it against the expected checksum resulting from the key exchange protocol.", "sentence2": "performing the gesture would require the user to move her position or even climb on to a table, which is obviously neither feasible to assume nor desirable.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2680", "sentence1": "Such a ges­ture is both difficult to recognize and perform.", "sentence2": "performing the gesture would require the user to move her po­sition or even climb on to a table, which is obviously neither feasible to assume nor desirable.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2681", "sentence1": "The physical setup of the experiment corresponds to the one shown in Figure 1.", "sentence2": "security proofs [20, 45] for SAS pro­tocols show that regardless of the computational power of X, the probability of its winning is 2-l, where l is the length of the checksum in bits.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2682", "sentence1": "Protractor performs worst of all algorithms, which seems to be due to the design of the gestures.", "sentence2": "the angles of the user's gestures are likely to vary significantly, causing Protractor difficulties in aligning them.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2683", "sentence1": "We teach the case study as an active learning exercise in class, although it could be used as homework.", "sentence2": "the semester's work was rated highly by the community and students.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2684", "sentence1": "Problem-Based Learning is good for technical evaluations, such as teaching sniffing, testing/configuring computers/networks, and secure coding.", "sentence2": "students also need to apply security within an organization.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2685", "sentence1": "In addition to publicly available logs, Attack Resilient Public-Key Infrastructure (ARPKI) allows certificates of multiple signatures to make them attack resilient [1].", "sentence2": "we propose an alternative approach to reduce the damage of a compromised CA key by imposing multiple signatures on a certificate, especially on server certificates.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2686", "sentence1": "The basic reasoning is that a certificate is valid only if it is published in an audit log.", "sentence2": "a domain owner must register a new certificate with an audit log in order to make this certificate valid.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2687", "sentence1": "It is simply speculative that it would provide the best defense at detecting and preventing the ARP attack used.", "sentence2": "it is possible to transfer HTTP traffic over a different port if specificed by the traffic origin.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2688", "sentence1": "There is also no direct rela­tion between the number of decisions that needs to be made and the total number of Activities an application has.", "sentence2": " Agarwal et al. [4] implemented a similar crowdsourcing scheme for iOS.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2689", "sentence1": "The write_buffer argument points to a buffer that encodes all the information related to the requested operations.", "sentence2": "the Request type field encodes the type of the request (BC_TRANSACTION, in this example), the InterfaceToken field specifies the remote service (com.android.internal.telephony.ISms), and the code field specifies which of its exported functions must be invoked (sendText).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2690", "sentence1": "In fact, the codebase of each applica­tion assumes to be executed within a very specific context that varies among different apps.", "sentence2": "failing to provide such context will inevitably lead the app to crash or behave incorrectly.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2691", "sentence1": "How to avoid detection may be studied in future work, however, it is known that avoiding detection of monitoring systems is typically an arms race.", "sentence2": "the possibility of integrating such mechanism with our approach represents one of the most promising and interesting future work direction.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2692", "sentence1": "To test the effectiveness of Njas, we first developed a simple application performing the sensitive operations that our current prototype is able to monitor and, if necessary, block.", "sentence2": "this application attempts to perform the following operations: initiate several network connec­tions, access sensitive files on the SD card (such as the user s photos), send text messages to premium numbers, and ac­cess the user s contact list.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2693", "sentence1": "Similarly, Njas patches the interactions with the services defined within an application.", "sentence2": "monitor needs to patch the Binder transactions related to the startSer­vice and the scheduleCreateService methods that are the analogous for services of the startActivity and schedule-LaunchActivity methods.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2694", "sentence1": "As a second part of our evaluation, we used our prototype to sandbox the execution of 20 real-world applications.", "sentence2": "we selected 7 real-world popular apps from the top-free lists of 7 different app categories on the Google Play market.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2695", "sentence1": "While this approach would be feasible from a theoretical point of view, it is a task that would require a significant engineering effort, and it would inevitably make the attack surface bigger.", "sentence2": "note that this is possible only for free apps, as the APK files of paid apps are stored in a non-readable location.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2696", "sentence1": "Independently to our work, a recent paper [5] proposes a sandboxing mechanism that does not require modifications of the Android system.", "sentence2": "this system requires the usage of an app granting all the possible Android per­missions and the reimplementation of many of the security checks usually implemented by the Android operating sys­tem.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2697", "sentence1": "Several papers analyze the current Android permission system [13, 27, 4].", "sentence2": "these works highlight the main problems that affect the current permission system (e.g., being too coarse-grained or not customizable by the user).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2698", "sentence1": "For this reason, monitor needs to patch the second Binder transaction as well.", "sentence2": "monitor patches the info argument, an instance of the Ac­tivityInfo class, which contains all the information on how an activity should be created.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2699", "sentence1": "A hidden, developmental feature called \"AppOps\", included with Android 4.3 and 4.4, allowed this to be done on a limited basis.", "sentence2": "this feature was removed in subsequent Android versions.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2700", "sentence1": "Our system requires the installation of a single compatible stub for each orig ap­plication.", "sentence2": "the same stub application can be used to enforce several different user-defined security policies.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2701", "sentence1": "Another area where Njas can enforce a fine-grained user-defined policy is related to the capability of sending text messages.", "sentence2": "the user might want to block sending text messages to a specific set of phone numbers.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2702", "sentence1": "In addition, Linux kernel is compact and gradually stablized, and hence discovering new kernel vulnerabilities would be a challenging task, if not impossible.", "sentence2": "vendor-specific kernels and device drivers would become the main targets of rooting apps in the future, as they are implemented by a single party and mostly closed source, without undergoing rigid security review.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2703", "sentence1": "Once the rootfs is mounted, the first Linux process init will be started to mount the rest of file systems from the devices, and then perform initialization procedures based on the configuration files (e.g., init.rc).", "sentence2": "some rooting traits are present on all rooted devices (e.g., su binary, SuperSU app), while others only on some.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2704", "sentence1": "That is, when the API has been hooked, the hooking function will be presented on the top of the call stack, in­stead of the hooked API (e.g., Runtime.exec).", "sentence2": "apps could determine whether the API call has been hooked by checking the function name on the top of an IO exception call stack.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2705", "sentence1": "For each hooked API, our analyzer logs the input parameters, and manipulates the output of the API attempting to evade the detection.", "sentence2": "we manually analyzed the source code to identify the root detection methods employed by these apps.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2706", "sentence1": "If all lists were equal or generation methods open, this method would be ac­ceptable.", "sentence2": "because each list is different and largely non-overlapping, the ability to alter results by the choice of list leaves the evaluation process open to manipulation, since an author can choose the list that offers the best agreement.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2707", "sentence1": "Phase-two domain-name-based in­dicators do not provide unique value to CND between 2.63% and 3.84% of the time.", "sentence2": "between 96.16% and 97.37% of domain-name-based indicators are uniquely provided by a single source.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2708", "sentence1": "A further difficulty with this situation is that there is no ready taxonomy or terminology for describing precisely what activity a malicious actor is performing.", "sentence2": "because each list is different and largely non-overlapping, the ability to alter results by the choice of list leaves the evaluation process open to manipulation, since an author can choose the list that offers the best agreement.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2709", "sentence1": "In practice, it is not possible to hold two of these three aspects fixed.", "sentence2": "list acquisition covers a con­secutive date range of 30 months with some core methods common to both phases.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2710", "sentence1": "These lists generally do not intersect.", "sentence2": "it appears that these lists do not converge on one set of malicious indicators.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2711", "sentence1": "If two organizations see events, we expect information sharing to be useful if those two events are somehow the same.", "sentence2": "various similarity measures between incidents may be equally valid and yet contradictory.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2712", "sentence1": "We propose that, when shared in bulk, even apparently low-value data without much structure, annotation, or context, can yield useful information.", "sentence2": "even organizations that have low resources or high security requirements should not be discouraged from participating in sharing communities.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2713", "sentence1": "A compelling feature of their work is that the initial data mining approach they propose can be accomplished using privacy preserving data mining techniques without a third party.", "sentence2": "one approach is to follow the principal of least disclosure , which states that systems should strive to disclose as little to others as possible, while still sharing [5].", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2714", "sentence1": "Cyber crime, increasingly the work of large networks of individuals, is best countered with a similarly organized structure [7].", "sentence2": "it clearly confirms that understanding and investigating cyber crime take a Network to Defeat a Network.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2715", "sentence1": "Benefits and specific solutions developed are discussed.", "sentence2": "it clearly confirms that understanding and investigating cyber crime take a Network to Defeat a Network.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2716", "sentence1": "This poses a threat not directly related to the sources themselves, but to the APIs to access them.", "sentence2": "an attacker can include a location provider reporting fake locations [24, 33], create a fake virtual GPS module [22], install an app providing bogus locations [14], or enable mock locations normally used for debugging purposes.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2717", "sentence1": "Anonymous credential systems have strong security properties.", "sentence2": "unforgeability in the sense that users cannot show credentials that they never obtained, and con­sistency in the sense that each credential belongs to a well­defined user such that multiple users cannot collaborate to obtain privileges that one user alone would not have got­ten.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2718", "sentence1": "In our mechanism, this property is achieved since information pertaining the mobile phone is only shared with the Service Provider in unintelligible form.", "sentence2": " the phone identifier included in the User Credential is carried over from the Phone Credential without revealing its value to the issuing Service Provider.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2719", "sentence1": "For distance-based verification, we use (LC , δC ) as Reference Area, where LC corresponds to the center of the U.S. state of Colorado, i.e., the location obtained from the latitude-longitude pair (39°, −105.55°), and dC to 371 920m (the distance from LC to each of the four state corners).", "sentence2": "(LC , dC ) corresponds to the smallest circular area that fully encompasses Colorado.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2720", "sentence1": "An extension to the Bitcoin protocol has been proposed for solving this problem7 .", "sentence2": "at the time of writing, the proposed extension was not yet accepted and deployed at full nodes.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2721", "sentence1": "As in our proposal, the authors build upon cryptocurrency systems like Bitcoin for avoiding the dependence on TTPs.", "sentence2": "the proposal is based on computation-intensive zero-knowledge proofs, making it less suited for more resource-constrained devices, and requires a TTP during the initial setup phase.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2722", "sentence1": "A genesis pseudonym is included in one of the outputs of a GPTx and might, dependent on the used access control approach, not be completely unlinkable to a user identity.", "sentence2": "for ensuring the unlinkability of pseudonyms and allowing unlinkable pseudonym changes (satisfying properties 1 and 4), we adapt state of the art techniques for anonymizing cryptocurrency transactions for realizing pseudonym mixing.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2723", "sentence1": "Upon compromise of the TTP, large-scale sybil attacks be­come possible and the trustworthiness of issued pseudonyms is greatly reduced.", "sentence2": "centralized pseudonym issuers be­come attractive targets for attacks, resulting in high opera­tional costs for maintaining their security.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2724", "sentence1": "To do so, we need to construct authA such that a is bound to a specific evaluation of P * on partial input a.", "sentence2": "this section describes basic system assumptions and the details of each phase.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2725", "sentence1": "However, this message is completely de-coupled from any of A's responses.", "sentence2": "no matter what it sends, B * learns no any additional information.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2726", "sentence1": "The main idea is that this get introduced approach might lead a user not just to new contacts but possibly to new com­panies or jobs, as recommended by their connections.", "sentence2": "a major goal for LinkedIn users is to discover and establish new connections based on criteria such as: common connec­tions, interests, membership and educational or professinal experience.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2727", "sentence1": "OSN Connection Spillover: ability to later establish actual OSN connections, based on prior off-line interaction that resulted in mutual agreement to connect.", "sentence2": "it should be possible for two peers to connect via the OSN at some point when they are online, if they have decided to do so as a result of sufficient degree of commonality among their profiles, e.g., at least k shared friends.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2728", "sentence1": "D2 Profile Privacy wrt Peers: ability to perform cer­tain OSN profile operations (e.g., compare respective sets of friends/connections) with mutual privacy.", "sentence2": "information learned from such operations must be limited to what is common to both peers.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2729", "sentence1": "The main conclusion is that K- Cloak is much worse than the other two, and Max-Ent is slightly better than Geo-Ind in terms of success prob­ability.", "sentence2": "geo-Ind was not specifically engineered against the same-origin attack, and there is no proof that Max-Ent, although intuitively appealing as a maximum­entropy construction, optimally delays the attacker under a constraint on the magnitude of the noise.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2730", "sentence1": "Additional steps such as chip separation and micro-bumps assignment [9] are necessary to generate a practical 2.5D IC placement.", "sentence2": "we neglect these steps in our placement algorithm since they do not affect the attack result.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2731", "sentence1": "In this section, we propose a privacy-preserving system using an algorithm from the same family as k-NN.", "sentence2": "k-NN is a specific case of kernel density esti­mation (as discussed in Section 2.2).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2732", "sentence1": "Under certain threat conditions, we can extend existing single data owner solutions to secure the multi-data owner scenario.", "sentence2": "now, wi ' is the encrypted basis for the i-th tuple s class, scaled by the kernel value.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2733", "sentence1": "Intuitively, the ability to observe plaintext query outputs can leak the classi.ca­tion of data tuples.", "sentence2": "the data host can ob­serve what tuples are used in the output construction, and the query output associates classi.cations to those tuples.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2734", "sentence1": "Note that our threat model allows for one of the cloud parties to pos­sess the querying role, so whichever party can observe the kernel value sums can conduct this attack.", "sentence2": "the proto­col must determine the classification without revealing the kernel value sums to any party.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2735", "sentence1": "The data owners may not be explicitly given querying permissions, say if the data owners and queriers are separate institutions.", "sentence2": "miscreants in each party may collude together, providing a joint alliance with both roles.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2736", "sentence1": "The DO-Q threat model covers any situation where a single party can possess both the data owner and querier roles.", "sentence2": " k-NN data is often sensitive in nature.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2737", "sentence1": "In figure 1, we show a snippet of output from our server s start up.", "sentence2": "using a statically configured key brings with it a number of issues, including insecurity and the inability to scale.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2738", "sentence1": "We observe a range of qualitative strategic behaviors, which vary in clear patterns across envi­ronmental conditions.", "sentence2": "we find that the efficacy of deterrent defense is critically sensitive to detection capa­bility, and in the absence of perfect detection the defender is often driven to proactive moving-target actions.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2739", "sentence1": "The forensics analyst uses the collector programs to dump the contents of RAM and flash memory.", "sentence2": "those collector programs are similar to the FTK imager and LiME forensics tools in terms of provided functionality.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2740", "sentence1": "An integer overflow is not exploitable directly as no buffers are overwritten.", "sentence2": "there is no actual memory overflow.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2741", "sentence1": "This observation is also evident by a growing underground market for OSN abuse, where accounts with a large number of connections are sold at a premium.", "sentence2": "we used the out-of-bag error estimates freely computed by the RF algorithm to numerically find the best number of decision trees and the number of features for each tree, so that the prediction variance and bias are controlled across the trees.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2742", "sentence1": "As shown in Figure 1, we performed parameter tuning to calibrate the RF classifier.", "sentence2": "we used the out-of-bag error estimates freely computed by the RF algorithm to numerically find the best number of decision trees and the number of features for each tree, so that the prediction variance and bias are controlled across the trees.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2743", "sentence1": "After that, we present background on defending OSNs against fake accounts.", "sentence2": "we divide fake account defense mechanisms into two categories, based on how new accounts are admitted into the OSN.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2744", "sentence1": "Assuming that fakes can establish only a small number of attack edges, the subgraph induced by the set of real accounts is loosely connected to fakes. ", "sentence2": "topology-based approaches rely on a key assumption that the cut crossing over attack edges is sparse, and accordingly, they aim to find such a sparse cut with formal guarantees [2, 52, 59].", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2745", "sentence1": "In the NTL library, the basic algorithm is used (with time complexity O(n 3 ) when all matrix dimensions are n).", "sentence2": " because the computations in the offline phase happen over random data, the overall problem is embarrassingly parallelizable, making it faster than existing solutions for processors with an appropriate number of cores.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2746", "sentence1": " For instance, physical devices provide a realistic setting for the test execution so allowing specific observations, e.g., battery consumption. ", "sentence2": "emulated devices are more flexible as they permit to simulate events and state changes, e.g., rotation, GPS coordinates changes or incoming SMS.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2747", "sentence1": "For instance, an application might attempt an illegal access by triggering a system component, e.g., consider a browser permitting to change the cache location.", "sentence2": "rA mon­itors the state of the entire file system and records every observed change.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2748", "sentence1": "These methods present outstanding results; however, the computation complex is high so that the power consumption of the target system significantly increases in some situations. ", "sentence2": "in this paper, we propose a secure energy consumption scheme, named the Secure Power Management (SPM for short) method, to control the power consumption of the WSN nodes by employing power management concept so that when they suffer malicious or insider attacks, the WSN nodes can still save power. ", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2749", "sentence1": "But it also causes higher performance penalty.", "sentence2": "because of the extensive use of the WSNs, such as healthcare monitoring [17] and industrial monitoring [18], most previous studies focused on data leakage prevention and error detection [19-21].", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2750", "sentence1": "Recently many studies have been proposed to monitor and detect the insider attacks.", "sentence2": "implementing an effective detection system is a very challenging task.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2751", "sentence1": "Any malware can behave like an APT as the attackers mainly used malware to identify, segregate and try to do data exfiltration while using different methodologies during a cyber-attack.", "sentence2": "quantifying the risk due to an APT is a big challenge as little has been understood about this security concern [3].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2752", "sentence1": "It is recommended that an SDN network employ switches from multiple man­ufacturers and companies.", "sentence2": "it also increases the probability that one or few switches in the network are compromised and attack the rest of the network infrastructure.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2753", "sentence1": " A compromised switch may not only lose its normal functionality, but it may also maliciously paralyze the network by creating network congestions or packet loss.", "sentence2": "it is important for the system to be able to detect and isolate malicious switches.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2754", "sentence1": "The switches report their status to the controller periodically, such as port statistics and flow statistics, according to their communication protocol.", "sentence2": "switches may contain vulnerabilities that can be exploited by attackers.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2755", "sentence1": "However, the detection rates reflect the previous conclusions regarding the limited applicability of each method (the algorithms detected distinct malicious switches, according to the related type of malicious behavior).", "sentence2": "in addition, each scenario consisted of 12 different cases in terms of the set of values for the following parameters: α,\nβ, and θ (see Table 1).", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2756", "sentence1": "In this work, we study the in­trinsic hardware characteristics of modern graphics process­ing units (GPUs) due to random manufacturing variations, and exploits the inherent randomness to generate device­specific signatures.", "sentence2": "we present a novel GPU­based hardware fingerprint scheme to generate a unique, sta­ble, physically unclonable, unpredictable, and random bit string from the inherent hardware features of a general pur­pose GPU (GPGPU).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2757", "sentence1": "Inside the SMX, we further evenly dis­tribute the number of cores into each execution cluster.", "sentence2": "two cores (when m is set as 128) are randomly selected from each execution cluster to produce the .nger­print in our investigated NVIDIA GPU Kepler Architecture.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2758", "sentence1": "The public key and certificate of the server are included in our application or  Figure 3: Comparative evaluation pre-installed in a displayer.", "sentence2": "memory attacks are a technique that reads and modi.es the memory of a target device using the vulnerability of application or framework of the device.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2759", "sentence1": "The public key and certificate of the server are included in our application or pre-installed in a displayer.", "sentence2": "attackers can't decrypt the encrypted messages in the polynomial time.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2760", "sentence1": "Unless attackers compromise both a displayer and a receiver at the same time, our method is secure against the memory attacks because the password doesn't exist on the device.", "sentence2": "our method is secure against key logging and memory attacks.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2761", "sentence1": "IoT applications need new cryptographic algorithms and protocols.", "sentence2": "the tiers have different processors, run different operating systems, and use different programming languages.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2762", "sentence1": "However, the accuracy was less than 0.9, except for Jan. - Mar. 2013.", "sentence2": "retraining the classifier is essential, but when to do so cannot be decided only by the results.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2763", "sentence1": "To detect malicious web pages used by drive-by download attacks, the paths of redirections from compromised web pages to attack­ers web pages are important, and divarications of redirec­tions are not important.", "sentence2": "we focus on the paths of redirections from the first accessed URL to the last accessed URLs.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2764", "sentence1": "To detect a huge variety of malicious web pages correctly, a large number of web pages is needed for training data.", "sentence2": "to prepare the training data of Prophiler, we used the discrimination results of web pages identified by high-­interaction honeyclients [1] as the labels of the content on web pages without manual inspections.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2765", "sentence1": "Programs such as the CubeSat Launch Initiative [7] in the United States and the European Space Agency's Fly Your Satellite [4] program encourage students and researchers to build and launch small satellites.", "sentence2": "in Section 3.5, it was required to choose 2 values from the initial array values.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2766", "sentence1": "To the best of our knowledge, these approaches do not consider all (possible) observations an adversary can make: in addition to performing first-last correlations, a network-level adversary can gain information by only observing parts of a circuit.", "sentence2": "our goal is to extend existing approaches by formally quantifying the advantage of a network-level adversary.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2767", "sentence1": "In this case, return-oriented programming (ROP) which is done by combining a chain of code fragments can be considered to circumvent the prevention [6].", "sentence2": "we can obtain that information through dmesg command.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2768", "sentence1": "Our method implements a four-step procedure to solve these problems.", "sentence2": "through the two steps above, we get a sequence of groups, which is consisted of a set of patch blocks and a source file related to them.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2769", "sentence1": "Each component of PIC service is detailed as the following: PI: PI should be well chosen to reveal the least privacy about Internet client.", "sentence2": "it should be the least rele­vant to the client's location, ISP, identity, and other sensitive characteristics.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2770", "sentence1": "The DNS authoritative servers can pose privacy threats because they are able to observe, collect, process, and transfer privacy-relevant DNS informa­tion (such as incoming DNS traffic) about clients either for the surveillance purposes of themselves or for facilitating an outside censor by passing him the data.", "sentence2": "the source IP address of client is of great interest to adversaries because it can be associated with the client s DNS messages to re.ect the end user s Internet activities.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2771", "sentence1": "While offline analysis remains unaffected by the new Android Runtime (ART), dynamic approaches which instrument the Dalvik Virtual Machine suffer from the fact that the interpreter is replaced by an ahead-of-time compiler suite.", "sentence2": " systems like TaintDroid [1] lack their instrumentation target and thus are not applicable anymore as of Android 5.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2772", "sentence1": "For Android versions up to 5.1, those permissions are requested and granted only once at installation time, while the approaching Android M release will feature dynamic permission revocation for a selected subset of permissions.", "sentence2": "the actual usage of those privacy-sensitive resources remains completely opaque to the user.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2773", "sentence1": "In our examples, data authentication processes walked through the complete authentication paths defined by the trust schema.", "sentence2": "note that even with key caching, the first authentication process may still involve several round trips of retrieving intermediate keys.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2774", "sentence1": "To explicitly link rules, we assign each rule a unique identifier to be used in a function-like way as part of the key name pattern, as shown on Figure 5(b).", "sentence2": " invoking such rules is similar to invoking a function: invocation substitutes the key name pattern with the data name pattern from the invoked rule, specializing it with the supplied pat-terns or references to the indexed sub-patterns.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2775", "sentence1": "Every successful authentication path must end at a trust anchor.", "sentence2": "a trust schema must always include a way for trust rules to establish the link(s) from data or key names down to a trust anchor.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2776", "sentence1": "We believe this syntax is sufficiently general to capture complex trust model frameworks, allowing reuse of trust models by different application instances.", "sentence2": "the trust schema for our blog example can be used by any other blog website that shares the same trust model.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2777", "sentence1": "To do so, we assume a key size of 1024b, batch size of 10, and signature size of 512KB.", "sentence2": "to help guide the decision about which variation to use, we make the following claims based on the application needs and assumptions.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2778", "sentence1": "Obfuscation is based on knowledge of the encryption key and the content name under IBAC protec­tion.", "sentence2": "even if an adversary knows the name N, it cannot generate N ' since it does not possess the ap­propriate key.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2779", "sentence1": "Thus far, we assumed that name encryption (obfuscation) keys are known to all authorized consumers in U(N).", "sentence2": "this might not be the case in practice.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2780", "sentence1": "Therefore, con­sumers must provide sufficient authentication information, e.g., via an interest signature.", "sentence2": "to implement autho­rized disclosure (in the presence of router caches), any en­tity serving content must (a) possess the information neces­sary to perform authentication and authorization checks and (b) actually verify the provided authentication information.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2781", "sentence1": "However, cached content needs to contain every authorization signature verification key that could be used to access said content.", "sentence2": "producers need to provide all possible public keys that can be used to access the content under IBAC protection.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2782", "sentence1": "For example, malicious insiders can harm their organizations by stealing intellectual property, sabotaging corporate IT systems, or misusing IT systems to execute financial fraud.", "sentence2": "in spite of rigorous researches, it still remains one of the most difficult challenges in the areas of security to prevent, detect, and respond to insider threats and attacks [2, 3].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2783", "sentence1": "As its name implies, the most striking attribute of GSPR is that it supports probabilistic revocation.", "sentence2": "gSPR's revo­cation check procedure does not produce deterministic re­sults, but instead produces probabilistic results, which may include false positive (i.e., false alarm) results but no false negative results.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2784", "sentence1": "This theorem prescribes that GSPR satisfies the anonymity property in the random oracle model when the DLIN as­sumption is presumed.", "sentence2": "gSPR's revo­cation check procedure does not produce deterministic re­sults, but instead produces probabilistic results, which may include false positive (i.e., false alarm) results but no false negative results.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2785", "sentence1": "This drawback can be mitigated in a number of ways, including the use of time-stamped parameters [16] or the use of accumulators [21].", "sentence2": "these methods incur additional overhead that may be unacceptable in many applications.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2786", "sentence1": "Specifically, we utilize the “Type A” internal described in pairing-based cryptography (PBC) library available at [1].", "sentence2": "to minimize the computational overhead, the verifier only runs RevCheck for a segments.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2787", "sentence1": "This mechanism is far from applicable within the environment of existing OSNs, as it leverages the multi-hop routing protocol of the specific OSN.", "sentence2": "our approach is designed for easy integration with existing social networks, relying on techno­logical capabilities widely available to such services.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2788", "sentence1": "The users that had selected a neutral stance, recog­nized the need that other users may have for preserving their privacy, but did not have a strong motivation in using such a mechanism.", "sentence2": "these users were also not aware of the true visibility of their photos.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2789", "sentence1": "Al Bouna et al. presented a system for preserving privacy regarding multimedia objects [21], which can be specifically used for photos [20].", "sentence2": "furthermore, the par­ticipants knew that the hidden users were friends of theirs.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2790", "sentence1": "The result of this phase is a processed photo that can be rendered selectively accord­ing to who is viewing it.", "sentence2": "when a photo is accessed, the system will automatically blur the faces of the users that have restricted access.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2791", "sentence1": "The enclave participates in attestation by invoking ereport, which generates a hardware-signed report of the enclave's measurement, and then sending the report to the verifying party.", "sentence2": "due to certain scalability challenges, we implement the following optimizations.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2792", "sentence1": "Furthermore, extending traditional type systems to enclave programs is non-trivial because the anal­ysis must faithfully model the semantics of SGX instruc­tions and infer information flows to individual addresses within enclave memory.", "sentence2": "we develop a static ver­ifier called Moat that analyzes the instruction-level behav­ior of the enclave binary program.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2793", "sentence1": "However, establishing such a guarantee requires precise understanding of the contract between the hardware and software.", "sentence2": "it requires specify­ing a formal API-level semantics of SGX instructions, an adversary model, and a verification methodology (with tool support) to detect insecure use of SGX instructions.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2794", "sentence1": "In FaceLive, the classification algorithm can be chosen from Bayesian Network, Binomial Logistic Regression, and Multilayer Perceptron.", "sentence2": "bayesian network (BN) is a probabilistic graphical model of joint multivariate probability distributions [13].", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2795", "sentence1": "Finally, the real-time response can be used as a liveness indicator in the assumption that legitimate users can interact with the system in real time while it is difficult for fake faces to do so.", "sentence2": "eye blink and head rotation are two typical real-time response based liveness indicators which have been used in popular face authentication systems such as Google's FaceUnlock [16].", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2796", "sentence1": "Figure 14 show that FaceLive achieves the EER rate in the range of [7.2%,7.5%] when the detection of a single type of facial land­marks is unsuccessful and the coordinate information of the facial landmarks is missing.", "sentence2": "the missing localization of nose has a slightly more significant impact on FaceLive than the missing localization of eyes Figure 13: Face photos in ideal lighting condition and under illumination in varied environments [18] and mouth.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2797", "sentence1": "When a user holds and moves a mobile device in front of a real face over a distance, the changes of the views of the face can be observed in the facial video.", "sentence2": "the horizontal movement of the device (along axis X of the device) leads to changes in yaw while the vertical movement of the device (along axis Y of the device) leads to changes in pitch.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2798", "sentence1": "In the third part, we record the inertial sensor data about the MFF-based attacks performed by participants.", "sentence2": "the participants are asked to perform the random-move attacks first and then the imitation-move attacks.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2799", "sentence1": "Finally, in Section 5 we provide an in-depth analysis of the 3 We stress that McOE-G, COPA and POET do not achieve full mis­use resistance, and only achieve a weaker notion called online authen­ticated encryption .", "sentence2": "it is immediate that nonce-based misuse-resistant encryption implies IV misuse-resistant en­cryption by simply using a random IV (and noting that for q encryptions the probability that an IV repeats is q 2/2£ where f is length of the IV).", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2800", "sentence1": "As previously mentioned, we also use this p-value computation in Stage 2 as a rough heuristic for decid­ing which disjunctions to keep and pass on to Stage 3.", "sentence2": "we stress that these Stage 2 p-value computations are not valid because the disjunctions are formed using the profiles from the training set, and hence are not independent of these same training set profiles.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2801", "sentence1": "To estimate regression coefficients in this model, we use a variant of Lasso called L1-regularized logistic regression [23], whose effectiveness has been established in several empirical studies across multiple domains (e.g., [5, 28]).", "sentence2": "they also show that favoring high preci­sion algorithms can yield better recall at high confidence, and that scaling output numbers may require to accept lower sta­tistical guarantees to find sufficient hypotheses.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2802", "sentence1": " The first algorithm is Set Intersection, which orders inputs by the fraction of profiles where the output is present they are covering. ", "sentence2": "the best candidates for targeted inputs are the one present in the largest fraction of the profiles with the output.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2803", "sentence1": "identification prior to (mutual) authentication is a problem that crops up in other systems as well, such as in e-passports [28] and RFID tags [29].", "sentence2": "3.2 SIM card The SIM is extended to store the new shared secret key κ next to two PMSIs; the currently active PMSI (PMSI) and the future PMSI (Pnew).", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2804", "sentence1": "Over time, the commercial IMSI catchers were extended with a lot of additional functionality such as eavesdropping on wireless calls.", "sentence2": "they are still, rather euphemistically, called IMSI catchers.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2805", "sentence1": "The data bearer throughput rapidly shrinks to zero when the downlink resource is captured by the signaling bearer.", "sentence2": "the data session cannot affect the throughput of the signaling bearer.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2806", "sentence1": "Compared with other spamming attacks [18,23,24], this threat readily bypasses the firewall and security boxes.", "sentence2": "the overhead of these schemes is definitely beyond that of the free XOR technique.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2807", "sentence1": "Moreover, the VoLTE signaling and voice bearers use the same IP allocated to the VoLTE interface, and the packets for these two bearers are differentiated based on their corresponding ports.", "sentence2": "only the packets with the RTP and RTCP ports of the session ID are delivered to the voice bearer, whereas the others are to the signaling bearer.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2808", "sentence1": "VoLTE is still at its early phase for global rollout.", "sentence2": "the network infrastructure also lacks proper access control and runtime check.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2809", "sentence1": "We discover several vulnerabilities in both its control-plane and dataplane functions, which can be exploited to disrupt both data and voice in operational networks.", "sentence2": "we find that the adversary can easily gain free data access, shut down continuing data access,or subdue an ongoing call, etc.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2810", "sentence1": "Ultimately, voice and data operate in the same, connection-less IP network.", "sentence2": "this paradigm shift is double-edged, exposing LTE networks and users to unanticipated vulnerabilities.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2811", "sentence1": "Since VoLTE continues to offer call service, it is natural to adopt the time-based charging, following the common practice for traditional CS voice.", "sentence2": "only the call duration on the data plane is collected for billing.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2812", "sentence1": "Moreover, some signals are even exchanged before the call is established, for example, SIP-INVITE, SIP-INVITE-OK, SIP-INVITE-ACK messages are used to set up a call.", "sentence2": "the practice to supply free VoLTE signaling does have loopholes.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2813", "sentence1": "VoLTE carries both its control signaling and voice message over IP packets.", "sentence2": "both planes can be tricked to transmit or receive non-VoLTE packets.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2814", "sentence1": "Reductions in garbled-circuit size [21, 22, 24]: Historically, the most expensive part of any secure protocol was the cryptographic operations.", "sentence2": "significant algorithmic improvements to secure protocols together with much faster implementations of cryptographic primitives (e.g., due to better hardware) have considerably changed the equation.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2815", "sentence1": "1.3 Our Results We construct fast garbling methods solely under the as­sumption that AES behaves like a pseudorandom function.", "sentence2": "medium circuits: In the larger SHA-256 circuit, where the majority of the gates are AND gates, there was a difference between the results in the two com­munication settings.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2816", "sentence1": "Recent re­search [15, 18] has shown that those mechanisms are indeed effective in increasing password entropy.", "sentence2": "advanced attackers are still finding ways to efficiently crack such pass­words offline using various forms of hybrid attacks (marry­ing dictionary attacks with brute-force attacks).", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2817", "sentence1": "Clients who care about privacy prefer to have their data encrypted on the client-side using semantically secure encryption schemes.", "sentence2": "naïve application of encryption thwarts deduplication since identical files are uploaded as completely independent ciphertexts.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2818", "sentence1": "We note that using Sphinx also for data forwarding would result in low throughput due to pro­hibitively expensive per-hop asymmetric cryptographic operations.", "sentence2": "we use Sphinx only for session setup packets, which are amortized over the subsequent data transmission packets.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2819", "sentence1": "Sessions are in particular not related to any long term secret or identifier of the host that creates them.", "sentence2": "two sessions from the same host are unlinkable, i.e., they are cryptographically indistinguishable from sessions of two different hosts.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2820", "sentence1": "To re­duce setup delay, HORNET uses only two setup packets within a single round trip between the source and the destination.", "sentence2": "session setup only incurs O(n) propagation delay in comparison to O(n 2) by the telescopic setup method used in Tor (where n is the number of anonymity nodes traversed on the path).", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2821", "sentence1": "Admittedly, the use of tagging, especially in conjunction with replay attacks, allows the adversary to improve the effective­ness of confirmation attacks.", "sentence2": "end-to-end MACs protect the integrity of the data, making such attacks (at a large scale) de­tectable by the end hosts.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2822", "sentence1": "Consequently, L3 Tor s and HORNET s goodput is 32% less than that of LAP and Dovetail.", "sentence2": "with the FSes for all nodes on both paths,FSf andFSb , S ii is ready to start the data transmission phase.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2823", "sentence1": "Furthermore, it is difficult for attackers to perform active attacks (e.g., packet replay) at scale while remaining undetected.", "sentence2": "our testbed contains an Intel software router connected to a Spirent TestCenter packet generator and analyzer [10].", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2824", "sentence1": "The experiments demonstrate the scalability and general­ity of the proposal.", "sentence2": "the experimental analysis reports evaluations on a wide range of password strengths, and of state-of-the-art attacks on very large datasets, includ­ing attacks that would have been prohibitively expensive to handle with existing simulation-based approaches.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2825", "sentence1": "Many passwords can only be guessed with significant effort, and motivated attackers may be willing to invest resources to obtain valuable passwords.", "sentence2": "it is eminently impractical for the defender to simulate expensive attacks against each user to accurately characterize their password strength.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2826", "sentence1": "However, in the case of DNS rerouting, this is achieved by hiding the origin’s IP address and relying on redirection through the use of the website’s domain name. ", "sentence2": "as illustrated in Figure 2, the website is only protected against trac that uses the domain name to initiate the connection.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2827", "sentence1": "When a web server initiates an outgoing connection on its own accord, the CBSP is not used as a proxy.", "sentence2": "the origin establishes a direct connection with an external host, effectively exposing its IP address to that particular host.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2828", "sentence1": "Addition­ally, the results of our large-scale analysis are lower bounds.", "sentence2": "as mentioned in Sec­tion 2.2, several CBSPs are CDNs that offer additional security services, and vice versa.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2829", "sentence1": "However, this certificate lists the domain name as the subject, and therefore identifies itself as the origin.", "sentence2": "if an attacker is able to scan all IP addresses and retrieve all SSL certificates, he can find the IP addresses of hosts with certificates that are associated with the domain he is trying to expose.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2830", "sentence1": "The most popular Cloud-based Security Providers do not require the purchase of dedicated traffic-rerouting hardware, but rely solely on changing the DNS settings of a domain name to reroute a website's traffic through their security infrastructure.", "sentence2": "this rerouting mechanism can be completely circumvented by directly attacking the website's hosting IP address.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2831", "sentence1": "This enables the CBSP to provide all the necessary DNS records in order for rerouting to take place.", "sentence2": "the configuration of additional custom records, such as the MX records to identify the domain's mail server, has to be managed through the CBSP's own custom interface where these additional records need to be added by the client.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2832", "sentence1": "For the domains under the protection of CloudFlare, the DNS records are managed by the CBSP.", "sentence2": "we excluded CloudFlare customers that were exposed through their A record, as this indicates that the administrator has deliberately paused the CBSP rerouting through CloudFlare's web interface.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2833", "sentence1": "Instead of storing the entire position map in the client s local memory, the client can store it in a smaller ORAM on the server.", "sentence2": "this position map ORAM needs to store N labels each of log N bits.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2834", "sentence1": "In an MPC scenario, the ORAM client s computation will be expressed as circuits that will be securely evaluated be­tween the multiple parties.", "sentence2": "an ORAM s circuit complexity is the total circuit size of the ORAM client algo­rithm over all rounds of interaction [53].", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2835", "sentence1": "The key insight is to maintain the invariant (Fact 2 in Appendix 9) that if some bucket in the real ORAM has an available slot, then during the post-processing of 1-ORAM, no blocks can be pushed through this bucket towards the root.", "sentence2": "to formalize this argument requires an intricate induction proof that is presented in Appendix 9.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2836", "sentence1": "If the limitation on the occupation of the physmap is for every active task, attackers can create many independent processes and each of them follows the restriction but however the total amount of the memory mapped by the physmap in these processes is enough to cover the SLAB caches storing the target vulnerable objects and the protection is bypassed by the attacker.", "sentence2": "the attention should be put on the memory usage of all the active tasks owned by every user logging on the system.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2837", "sentence1": "And it needs extra memory for spraying vulnerable objects and making them dispersedly allocated in kernel space.", "sentence2": "attackers can operate on this file descriptor in a user program and make the kernel reuse the freed PING sock object, which leads to code execution in the kernel.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2838", "sentence1": "Functional encryption can be used to implement audit over encrypted logs: The declas­sification function can perform the audit and return the outcome.", "sentence2": "cl initiates the audit process by choosing a policy.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2839", "sentence1": "No other information about the log can be recovered by an adversary looking at the log's encryption.", "sentence2": "a log's encryption reveals neither the actual values on the log (other than constants occurring in the policy), nor the equality between values in non-joinable columns.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2840", "sentence1": "The general idea is that the neighborhood can be formulated by some constraints of data or distance (metric) functions instead of adding or removing a record.", "sentence2": "applying these neighborhood based differential privacy is not feasible in our model because there is only one sole tuple (location) at each timestamp without any \"neighbors \".", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2841", "sentence1": "Figures 10c and 10d show the precision drops when k ' rises because of a larger returned set.", "sentence2": "figures 10e and 10f indi­ cate recall increases with large k ' .", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2842", "sentence1": "LPPMs generally use obfuscation methods, such as spatial cloaking, cell merging, location precision reduction or dummy cells, to achieve anonymity based privacy or uncertainty based privacy.", "sentence2": "anonymity or ad hoc uncertainty based techniques do not always provide sufficient privacy protection [22, 37].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2843", "sentence1": "The hardware virtualization makes it possible to execute most of the malware instructions as native CPU instructions on the real hardware without any interception.", "sentence2": "it does not suffer from inaccurate or incomplete sys­tem emulation.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2844", "sentence1": "During the alignment process, two meta-nodes are recursively processed to compute the similarity score.", "sentence2": "to compute the similarity score between two meta-nodes, we first perform sequence alignment of the sequences corresponding to the meta-nodes.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2845", "sentence1": "This approach is e.ective when large portions of the sequences have unique alphabets, such as the lines of a source code.", "sentence2": "a system call sequence has a limited alphabet, while the sequence itself is usually long.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2846", "sentence1": "Potentially, malware can have multiple evasion points.", "sentence2": "a malware can perform multiple evasion checks at different sections of the system call sequence.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2847", "sentence1": "In this result, we can see that with ω > 83 we achieved 100% recall rate.", "sentence2": "all evasion calls of the ground truth dataset are captured when ω > 83.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2848", "sentence1": "The afine gap penalty combines both constant and linear gap penalties, taking the form ga + (gb * L).", "sentence2": "it assigns an opening gap penalty ga, which increases with the rate of gb.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2849", "sentence1": "The Power of Procrastination: Detection and Mitigation of Execution-stalling Malicious Code.", "sentence2": "the clustering of evasion signatures as described in Section 4.4 can help improve this manual process.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2850", "sentence1": "This is a simple and fast schema.", "sentence2": "this schema gives too much freedom for sequence alignment, resulting in unnecessary long gaps.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2851", "sentence1": "Different tools are available to help malware analysts in this process.", "sentence2": "these tools in practice require considerable manual input along with auxiliary information.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2852", "sentence1": "Automated dynamic malware analysis is a common approach for detecting malicious software.", "sentence2": "many malware samples identify the presence of the analysis environment and evade detection by not performing any malicious ac­tivity.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2853", "sentence1": "On Linux, signal handlers are usually not invoked by any application code5, so they do not return to the application code.", "sentence2": "as mentioned by the authors of COOP, CFI solutions that generate fine-grained CFGs based on class hierarchies tend to be immune to COOP.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2854", "sentence1": "However, since addr has already been activated, the current ECFG allows an indirect branch to target addr through newly added edges.", "sentence2": "address activation accommodates dynamic linking.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2855", "sentence1": "In addition, each pCFI-protected application runs with instrumented libraries.", "sentence2": "we also modi.ed and instrumented standard C/C++ libraries, including the musl libc, libc++, libc++abi, and libunwind.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2856", "sentence1": "At runtime, the program adds edges on the .y, but pCFI disallows addition of any edge not included in the static, all-input CFG.", "sentence2": "the all-input CFG serves as the upper bound for what edges can be added to the enforced CFG during runtime.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2857", "sentence1": "When we merge all benchmark's results, about 30% of indirect branch targets are activated in total, slightly more than the result triggered by Octane 2.", "sentence2": "as can be seen, the address activation shows steep growth even at the end; on the other hand, it does not activate more target addresses com­pared to other input data sets, which trigger similar ECFG growth as 400.perlbench.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2858", "sentence1": "Besides, functions' addresses can also be explicitly taken at runtime by the libc function dlsym.", "sentence2": "we changed dlsym's implementation so that before dlsym returns a valid function address, a pCFI runtime trampoline is called to activate the function's address.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2859", "sentence1": "Coarse­grained CFI (e.g., CCFIR [34], binCFI [35], and kBouncer [21]) enforces a coarse-grained CFG in which there are a few equiva­lence classes of target addresses (or even just one equivalence class) and an indirect branch is allowed to target one of the equivalence classes.", "sentence2": "pCFI starts executing a program with the empty CFG and lets the program itself lazily add edges to the enforced CFG if such edges are required for the concrete input.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2860", "sentence1": "We consider an exception han­dler s address activated when the function where the exception han­dler resides gets executed for the first time.", "sentence2": "same as how pCFI instruments and patches C++ constructors, pCFI also instru­ments those functions that have exception handlers when loading the code into memory and patches the code back to its original bytes when such functions are executed for the first time.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2861", "sentence1": "An idempotent operation is designed so that the effect of performing it twice is the same as the effect of performing it only once.", "sentence2": "after the first time, there is no need to perform it again.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2862", "sentence1": "For code-injection attacks, pCFI enforces DEP (Data Execution Prevention) at all times; its runtime enforces this by intercepting and checking all systems calls that may change memory protection, including mmap, mprotect and munmap.", "sentence2": "code injection attacks are impossible for programs that do not generate code at runtime.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2863", "sentence1": "This can be achieved by code instrumentation; that is, by inserting extra code that adds edges into the original program.", "sentence2": "such instrumentation can be costly since every time the indirect branch gets executed, the edge­addition operation needs to be performed.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2864", "sentence1": "[29] proposed an approach to .ne-grained CFI with (partial) modularity support.", "sentence2": "it does not protect return instructions, and its modularity support in­troduces time windows for attacks during dynamic module linking.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2865", "sentence1": "This paper considers side-channel attacks exploiting statistics values exported by procfs from co-located applications running within the same OS.", "sentence2": "we consider the default settings of procfs, which do not restrict accesses to a process' private directories in procfs by other processes.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2866", "sentence1": "The second line of research generalizes the definition of differential privacy for statistical databases.", "sentence2": "chatzikokolakis et al. [13] broadened the definition of differential privacy by parameterizing the definition with a distance metric d, and requiring that the degree of indistinguishability of two databases be a function of their distance.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2867", "sentence1": "Moreover, the domain of storage side-channel defense introduces important differences that require innovation.", "sentence2": "ince the values that our system must perturb to interfere with side channels are ones that are used by other software, it is important that our modifications do not violate invariants on which that software depends.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2868", "sentence1": "IFCC may limit the number of available ACICS, but it cannot prevent the Control Jujutsu attack in general.", "sentence2": "using our ACICS discovery tool, we were able to easily expand on our original exploit for Apache and develop an additional full exploit based on an ACICS with an arity that matches its ICS with its target function.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2869", "sentence1": "The liveness of an ACICS allows us to reason about its exploitability; if the liveness persists across the program lifecycle, the ICS can be attacked by almost any memory read/write vulnerability, regardless of where it occurs temporarily.", "sentence2": "a context-insensitive analysis, in contrast, does not distinguish between different invocations of a function, i.e., analysis results for local variables, arguments, and the return values from different invocations of the function are merged together.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2870", "sentence1": "This is in fact an O(n) operation if full anonymity in the set is desired, as any row skipped by the server leaks information about the query.", "sentence2": "the cost of computation may be traded off by being anonymous in a proportionally smaller set of the users.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2871", "sentence1": "We show the impact of the implementation on the speed of poly­nomial multiplication on Section 6.", "sentence2": "the Fast Fourier Transform strategy outlined in Section 5.4 significantly reduces the evaluation time per column (over an order of magnitude gain).", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2872", "sentence1": "Specififically, an app may query another app's content provider by directly referring to its authority, one or more URIs formatted in a Java­style naming convention: e.g., com.example.provider.ima geprovider.", "sentence2": "just like what happens to other attributes, such a reference (to the authority) can also become hanging, when the related provider is in absence on a device.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2873", "sentence1": "GUITAR must rebuild the recovered objects function dispatch tables to al­low the windowing system to invoke any inherited drawing functions.", "sentence2": "due to lack of type and symbol infor­ 2This mapping is done using a newly started stub pro­cess in the emulator, before any heap or data segments are allocated, to avoid con.icts with new live memory usage.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2874", "sentence1": "This login is cached and a timer is used to automatically log the user out after some time of inactivity.", "sentence2": "if someone later opened the app it would again ask for login credentials.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2875", "sentence1": "GUITAR requires no modification to the Android frame­work code but leverages the (open-source) data structure definitions of its windowing system.", "sentence2": "android always tries to save memory, and when an app is backgrounded its GUI tree will be deallocated and critical pointers within it (in particular the ones from TreeN­odes to their DrawOpLists) will be set to NULL (a good pro­gramming practice, but bad for memory forensics).", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2876", "sentence1": "A single drawing function may create mul­tiple draw ops and store them in one or more leaves.", "sentence2": "whenever the app is visible, a GUI tree of parent TreeNodes (describing relative geometric positions and screen layout hierarchy) and leaf TreeNodes (containing actual graphical draw ops) will be created in the process heap memory.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2877", "sentence1": "However, each leaf only describes its coordinates relative to its parent TreeNode.", "sentence2": "to find the neighboring leaves and compute each leaf's true (full screen) coordinates, GUITAR must look backwards through the tree's hierarchy.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2878", "sentence1": "When we performed the LG G3 s WhatsApp experiment, the last GUI we viewed was the app s Friends List window.", "sentence2": "finally, GUITAR recreates the runtime environ­ment to redraw the GUI using an unmodi.ed Android win­dowing system binary, and outputs the app s redrawn GUI as it would have appeared had it been displayed on-screen when the memory image was taken.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2879", "sentence1": "We therefore need to strengthen the previous composi­tion result for multi-stage key exchange protocols [17] to cover, first, key exchange sessions and stages which are only unilaterally authenticated or completely unauthenticated, and, second, protocols that do not allow for a public session matching, but for one where session partnering at a certain stage i is deducible given all stage-j keys for j < i. Jump­ing ahead, knowledge of earlier stages keys can be taken for granted as such keys can be revealed without impairing the chances of winning in a key-independent setting, which is in any case required for composition.", "sentence2": "as both achieve key independence, the analyzed TLS 1.3 handshake drafts are amenable to our composition result.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2880", "sentence1": "They do not consider the draft-dh draft, nor do they cover the resumption step.", "sentence2": "our approach of integrating resumption via composition may also be viable for their model.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2881", "sentence1": "The scale of the window may vary from thousands to millions of instructions.", "sentence2": "mimicry attacks on host-based intrusion detection systems.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2882", "sentence1": "Besides normal program behaviors ground truth (Sect. 6.1), we generate four types of synthetic aberrant path anomalies.", "sentence2": "small n (local feature analysis) makes it possible for attack­ers to evade the detection by constructing a malicious trace of which any small fragment is normal.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2883", "sentence1": "This approach detects aberrant path attacks because long n-grams are large execution windows.", "sentence2": "it re­sults in exponential training convergence complexity and storage complexity.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2884", "sentence1": "Malware clas­si.cation aims at extracting abstract malware behavior sig­natures and identi.es a piece of malware using one or multi­ple signatures.", "sentence2": "program anomaly detection mod­els normal behaviors and exams an entire profile to decide whether it is normal.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2885", "sentence1": "The essence of n-gram is to model and analyze local fea­tures of program traces with a small n. Enlarging n results in exponential convergence and storage issues.", "sentence2": "small n (local feature analysis) makes it possible for attack­ers to evade the detection by constructing a malicious trace of which any small fragment is normal.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2886", "sentence1": "How­ever, the underlying detection paradigms restrict the solu­tions from correlating arbitrary events in a long trace.", "sentence2": "their solutions do not detect general aberrant path attacks.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2887", "sentence1": "To reconstruct web-based attacks, forensic analysts typically rely on browser cache files and system logs.", "sentence2": "cache files and logs provide only sparse information often lacking adequate detail to reconstruct a precise view of the incident.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2888", "sentence1": "Lastly, it replaces the value of the Platform::current() pointer with the address of the newly created PlatformWrapper.", "sentence2": "as shown by the results reported in Section 6, WebCapsule performed remarkably well on a large variety of sites, considering that our implementation is an academic-level prototype system.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2889", "sentence1": "At this point, the analyst would need to explore the detailed browsing his­tory of these users, in an attempt to reconstruct how the credentials were actually leaked and learn how the phishing attack unfolded.", "sentence2": "permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for pro.t or commercial advantage and that copies bear this notice and the full cita­tion on the first page.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2890", "sentence1": "For example, when JavaScript Date() objects are instantiated to re­trieve the current system time, V8 calls OS::TimeCurrent-Millis() (via a call to RuntimeDateCurrentTime).", "sentence2": "using this technique, we are able to return the desired network response to Blink.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2891", "sentence1": "In this process model, WebCapsule could record multiple tabs independently.", "sentence2": "all our results refer to ex­periments performed on one single tab.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2892", "sentence1": "These benchmarks include Web Standards Project's Acid3 [1, 32] and the Dromaeo Test Suite developed by Mozilla [10].", "sentence2": "this is not an option for us, because it would violate our main goal of not altering any code outside of Blink (to completely inherit its portability) and of introducing only low overhead so that WebCapsule can be used as an \"always on\" forensic data collection system.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2893", "sentence1": "This is crucial to allow for a secure migration from an existing in­dexing system based on unique uidi s to our pseudonymous system.", "sentence2": "such a verification must be explicitly al­lowed by the converter.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2894", "sentence1": "A dual-mode signature scheme consists of the algorithms (SigKGenG, SignG, EncSignG, DecSignG, VfG and also uses an encryption scheme (EncKGenG, EncG, DecG) that has the group G as message space.", "sentence2": "encSignG(ssk, epk, C) : On input a signing key ssk, a public encryption key epk, and ciphertext C = EncG(epkX , EncG(epkB , xnymi,A))}(sid, qid).", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2895", "sentence1": "For instance for the use case of a social security system, it might be allowed that the health care provider can request data from the tax au­thority but should not be able to access the criminal records of its registered users.", "sentence2": "there is no need to learn for which particular user a request is made, or whether several requests belong together.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2896", "sentence1": "Pseudonymization of patient identifiers for translational research.", "sentence2": "current solutions require the central authority to be a fully trusted party, as otherwise it can provide false conversions and ex­ploit the data it learns from the requests.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2897", "sentence1": "An alternative approach is to use domain-specific pseudonyms, where only a central authority knows the cross­domain relation between the pseudonyms.", "sentence2": "current solutions require the central authority to be a fully trusted party, as otherwise it can provide false conversions and ex­ploit the data it learns from the requests.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2898", "sentence1": "The tricky part is to make this whole pseudonym gener­ation and conversion process verifiable and consistent, but without harming the unlinkability and blindness properties.", "sentence2": "for pseudonym generation a server SA must be ensured that it receives correctly formed pseudonyms.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2899", "sentence1": "A dual-mode signature scheme consists of the algorithms (SigKGenG, SignG, EncSignG, DecSignG, VfG and also uses an encryption scheme (EncKGenG, EncG, DecG) that has the group G as message space.", "sentence2": "the algorithms working with encrypted messages or signatures also get the $ keys (epk, esk) .EncKGenG(1t ) of the encryption scheme as input.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2900", "sentence1": "It can of course be realized with a generic multiparty protocol, where the first server SA inputs the pseudonym to be con­verted and the converter inputs all its secret keys, and the output of the second server SB would be the converted pseudonym, provided that the input by SA was a indeed a valid pseudonym.", "sentence2": "such a computation would be rather inefficient.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2901", "sentence1": "In a similar vein, it would be desirable to combine our pseudonym system with policy enforcement tools in a privacy-preserving manner.", "sentence2": "allowing the user to specify which data exchanges are permitted and enable the converter to blindly check whether a received conversion request violates any user constraints.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2902", "sentence1": "We validated classification results with 10-fold cross-validation.", "sentence2": "next, we discovered that the ratio of unreachable retweet­ers led to more errors than other features in CrowdTarget.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2903", "sentence1": "We observe that the crowdturf­ing and normal account groups have the same pattern: low retweet similarities.", "sentence2": "the black-market account group has the highest retweet similarity.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2904", "sentence1": "From the results, we confirm that most crowdturning and black­market accounts perform retweets without clicking on con­tained links because they have no reason to visit the links to retweet them.", "sentence2": "we use the click information as the final feature of crowdturng tweets.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2905", "sentence1": "Further, the smallest standard deviation of the black-market tweets shows that most of them are retweeted around the mean time.", "sentence2": "we decide to use the standard deviation of a retweet time distribution as a feature.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2906", "sentence1": "Most of the black-market tweets collected have negative skewness, implying that the num­ber of retweets gradually increases at first, but suddenly decreases later.", "sentence2": "c finally pays virtual money for the tasks that W has conducted.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2907", "sentence1": "In contrast, the black-market account group has the highest retweet similarity.", "sentence2": "on average, we paid $5.6 for 100 retweets and $13.4 for 1,000 retweets.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2908", "sentence1": "Malicious crowdsourcing, also known as crowdturfing, has become an important security problem.", "sentence2": "detecting accounts performing crowdturfing tasks is challenging because human workers manage the crowdturfing accounts such that their characteristics are similar with the characteristics of normal accounts.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2909", "sentence1": "We manually analyzed the normal tweets classified as ma­licious by CrowdTarget (i.e., false positives).", "sentence2": "figure 5b shows the standard deviation of the retweet time distribution of the three groups.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2910", "sentence1": "We note that in [26] and [7], the offline stage is independent of the circuit being evaluated in the online stage, whereas in our protocol, a single circuit is fixed for all computations.", "sentence2": "they are better suited for settings in which the function to be computed is not known ahead of times.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2911", "sentence1": "The only modification needed is to commit on all the commitments of the input labels using a trapdoor commitment, so that in the simulation in case P1 is corrupted, the simulator could change the commitments on the input labels after it learns P1’s input. ", "sentence2": "(Note that once we model H as a random oracle, we can also use it as the statistically-hiding extractable commitment.)", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2912", "sentence1": "If P2 asks to decommit (i.e., b = 1) then P1 simply decommits using the standard (canonical) decommiment and P2 checks that the XORs in all split commitments to a value are the same.", "sentence2": "if b = 0, then P1 sends P2 the XORs of the split commitment values.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2913", "sentence1": "We use a scoring mechanism to improve the success rate of matching process.", "sentence2": "algorithm 3 is leveraged to complete such process, it takes the predicted profiles as input, and then scores the similarity between the predicted profiles and the words in the dictionary with the same length.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2914", "sentence1": "On one hand, smaller k clusters less neighbors, reducing the chance of finding the right label.", "sentence2": "larger k associates more labels to a movement, making the correct label less distinguishable.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2915", "sentence1": "In other words, only all the results of the 6 classifications are non-zero will lead to a non-zero final result.", "sentence2": "k in this attack could not be determined by only testing a single classification.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2916", "sentence1": "8.2 Emanation-based Attacks The emanations produced by electrical devices were known to leak information regarding users activities.", "sentence2": "various emanation attacks that aim to infer keystrokes have been proposed and proved to be effective.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2917", "sentence1": "The standard k-NN algorithm produces only one label for each input, which reduces the chance of success for adversary, since only all the 6 successive classifications were made correct de­cisions would result in the correct PIN code.", "sentence2": "we mod­ify the standard k-NN algorithm to output all the possible Labels with their corresponding probabilities.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2918", "sentence1": "However, the accuracy of our attack depends on all the 6 successive classifications.", "sentence2": "only all the results of the 6 classifications are non-zero will lead to a non-zero final result.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2919", "sentence1": "Expectedly, the success rate should increase and we evaluate this hypothesis through a modified version of the prior experiment.", "sentence2": "we constructed a dictionary using 4 news reports [4, 15, 9, 7] from BBC News, all were writ­ ten by Smitha Mundasad, a health reporter specialized in medical 7We replace the word obfuscating with obfuscation , since we could not .nd the word obfuscating in the dictionary.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2920", "sentence1": "In this paper, we make the first step of exploring such threat space, and we show it is feasible to infer user’s highly sensitive information, e.g., PINs and typed texts, through data collected from the built-in sensors, including accelerometer and microphone. ", "sentence2": "we propose a set of new techniques to model user’s hand movement and reduce the interference from noises, and our attack is able to achieve high accuracy in keystroke inference. ", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2921", "sentence1": "In contrast, software-based attestation [24, 29, 47, 48] require neither secure hardware nor cryptographic secrets.", "sentence2": "security guarantees of software-based attestation methods rely on strong assumptions, such as the adversary being passive while the attestation protocol is executed and optimality of the attestation algorithm and its implementation.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2922", "sentence1": "This means that, although the adversary, denoted as ADV, can manipulate the software of (i.e., compromise) any device D in S, it cannot physically tamper with any device.", "sentence2": "aDV can eavesdrop on, and manipulate, all messages between devices, as well as between devices and VRF.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2923", "sentence1": "Whereas the 32-bit x86 implementation uses memory seg­mentation to isolate the safe region, the fastest 64-bit x86 implementation uses information hiding.", "sentence2": "it turns out that the hidden safe region was sufficiently large to be located and parsed using a modified version of the memory disclosure attack by Siebert et al.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2924", "sentence1": "With our proposed fix, IFCC uses three more instructions to validate each target which brings the total number of added instructions up to 15 per indirect call.", "sentence2": "we assume the presence of an adequate shadow stack im­plementation.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2925", "sentence1": "This allows it to prevent access and leakage of the shadow stack address.", "sentence2": "these security guarantees come with an average run time overhead of 19% which is considered impractical.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2926", "sentence1": "When a function (the caller) calls another function (the callee), the callee cannot determine which of the caller s registers are used at the moment of the call.", "sentence2": "the callee saves all registers it needs to use during its execution temporarily on the stack.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2927", "sentence1": "On the other hand, isolating the stack can potentially mitigate our attacks.", "sentence2": "once this address is known, the program can perform PC-relative references.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2928", "sentence1": "In principle, both CFI schemes provide strong protection, and even resist the latest code-reuse attack techniques such as COOP (counter­feit object-oriented programming) [40].", "sentence2": "the im­plemented CFI checks contain weaknesses due to unanticipated interactions with optimizations applied by the com­piler.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2929", "sentence1": "There is no clear definition in the literature with respect to the terms fine and coarse-grained CFI policy.", "sentence2": "the general understanding of a fine-grained CFI policy is that only branches intended by the programmer are allowed.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2930", "sentence1": "The target program contains a memory-corruption vulnerability that allows the adversary to launch a run-time exploit.", "sentence2": "we focus on vulnerabilities that allow the adversary to read (information disclosure) and write arbitrary memory.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2931", "sentence1": "As a result, active tags can be equipped with computational platforms that run crypto­graphic and security protocols to mitigate cloning attacks [5].", "sentence2": "passive tags do not enjoy these luxuries and there­fore are more prone to cloning attacks.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2932", "sentence1": "The rest of this paper is organized as follows.", "sentence2": "this may not always be the case.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2933", "sentence1": "Due to the existence of noise, no attacker whether using software or hardware to counter the physical challenges issued by PyCRA can instantaneously detect the change in the physical challenge.", "sentence2": "there always exists a non-zero probability of the attacker missing the changes in the physical challenge.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2934", "sentence1": "The primary goal of an attack is to replace the true physical signal that a sensor aims to sense with a malicious signal.", "sentence2": "an adversary will attempt to \"inject\" a signal into the physical medium that the sensor is measuring in order to jam or spoof the sensor.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2935", "sentence1": "Observe that we have not completed the run for A4.", "sentence2": "we know by our previous analysis that if A2 is successful, so will A4 (if we complete the run of A4).", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2936", "sentence1": "We note that whenever the communication partners exchange certificates in HMQV to authenticate their public keys this not only affects the message size but also the key derivation.", "sentence2": "the time for key derivation is increased by the time for the additional certificate verification.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2937", "sentence1": "All participants learn the model and thus can use it locally and privately, without any communication with other participants and without revealing the input data or the model's output to anyone.", "sentence2": "in contrast to conventional deep learning, there is absolutely no leakage while using the model.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2938", "sentence1": "The scenarios we consider - for example, collaborative learning of image recognition models between medical institutions - involve participants who are not actively malicious.", "sentence2": "it is reasonable to assume a \"passive\" adversary model, in which the participants execute the protocol as designed but may attempt to learn or infer sensitive information from other participants' data.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2939", "sentence1": "Our approach is independent of the specific algorithm used to construct a model for a particular task.", "sentence2": "it can easily accommodate future advances in neural­network training without changing the core protocols.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2940", "sentence1": "To provide secure and efficient communication among components, SeCage follows the idea of separating control plane from data plane by using the VMFUNC feature from Intel's hardware­assisted virtualization support.", "sentence2": "seCage first designates the security policy on which secret compartment is invokable by another compartment to the CPU, and allows such an invocation from being done without the hypervisor intervention.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2941", "sentence1": "When TrustOTP is running, the Rich OS will be suspended until TrustOTP exits.", "sentence2": "if the OTPs are being displayed for a long time, the user cannot perform any operations in the Rich OS.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2942", "sentence1": "When the system boots up, the code and data of the OTP generator will be loaded into a secure memory region that can only be accessed in the secure domain.", "sentence2": "the Rich OS cannot access the sensitive information directly from the memory.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2943", "sentence1": "It is flexible to support various OTP algorithms and multiple OTP instances on one smart­phone.", "sentence2": "besides protecting the OTP code integrity and the seed secrecy, we also need to protect the timer and the counter that are used as input to TOTP and HOTP, respectively.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2944", "sentence1": "It is reasonable because the same operation takes more time to complete in the Rich OS when TrustOTP is running.", "sentence2": "the performance of CPU and RAM decrease more than that of the peripherals.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2945", "sentence1": "Due to the OTP standardization work, most OTP instances use the same algorithms but different seeds and other parameters such as counters.", "sentence2": "we can support a large number of OTP instances in our system.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2946", "sentence1": "The interrupt handler of the NMI is in the secure domain, so the Rich OS cannot manipulate the handler either.", "sentence2": "we can guarantee that the system will switch into the secure domain as soon as the NMI is triggered.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2947", "sentence1": "An alternative method to achieve a secure display is to reuse the Rich OS's IPU driver after verifying its integrity.", "sentence2": "it cannot guarantee to provide a reliable display for OTP when the Rich OS crashes.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2948", "sentence1": "However, in contrast to the re-registration attack which involved addition and deletion of Sybil identities, the attacker does not fully control the process of obtaining new attack edges, since an honest user must accept or interact with a Sybil identity to establish a new edge.", "sentence2": "careful attention is needed to model the capabilities of an attacker aiming to induce and exploit churn in attack edges (see Sec­tion 3.2 for a formal description of our dynamic attack edge model ).", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2949", "sentence1": "If y marks the communication as unwanted, the balance B of (x, y) is lowered by one (from the sender s perspective).", "sentence2": "sybilLimit is vul­nerable to the re-registration attack discussed previously, since it allows the initiator of the random route to overwrite the public key registered at its tail.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2950", "sentence1": "If z marks the communication as unwanted, the credit balance of both edge (x, y) and edge (y, z) will be reduced by one.", "sentence2": "if the victim z has less edges compared to the number of attack edges owned by the attacker, a user targeting attack becomes practical.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2951", "sentence1": "SumUp assumes that the number of the attack edges is bounded.", "sentence2": "after examination, we find that the security guarantee of SumUp heavily relies on another underlying assumption, which is that the vote collector is placed far away from the attack edges.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2952", "sentence1": "In this section, we first introduce the SybilLimit proto­col [35], and then present our novel temporal attacks that allow an adversary to break SybilLimit s security guaran­tees.", "sentence2": "we show that an adversary can eventu­ally register an unbounded number of Sybil identities in the SybilLimit protocol (at a single instant in time).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2953", "sentence1": "We first leveraged existing compiler optimizations to reduce the usage of push/pop.", "sentence2": "modern compilers like gcc in most scenar­ios prefer using mov instruction for performance gain, as CPU can do pipeline scheduling for mov instructions.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2954", "sentence1": "We enforced the alignment by leveraging gcc's supports for multiple platforms.", "sentence2": "on platforms like ARM, function invocations do not automatically save the return address on stack, so gcc already understands how to align stack when there is no return address on stack.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2955", "sentence1": "ASLR-GUARD can either prevent code pointer leaks or render their leaks harmless.", "sentence2": "aSLR-GUARD makes it impossible to over­write code pointers with values that point to or will hijack the control flow to a desired address when the code pointers are dereferenced.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2956", "sentence1": "Type-based CFIs [2, 24, 31, 42] reduce the average indirect targets (AIT) by checking indirect branches with type.", "sentence2": "mCFI [31] checks type and number of parameters, Forward-edge CFI [42] checks the number of parameters, Safedispatch [24] and vfGuard [2] check class type for virtual function calls.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2957", "sentence1": "Next, we instrument the target program to correctly decrypt code locators.", "sentence2": "we first instrument every indirect call with code address in register in the ways shown in Table 3.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2958", "sentence1": "In Linux, multi-thread feature is enabled with Native POSIX Thread Library (NPTL), which creates a new thread with the clone system call.", "sentence2": "to create new thread, NPTL first allocates a new stack and stores thread descriptor at the end of the new stack.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2959", "sentence1": "Semantically secure encryption [12] guarantees that every encryption of the same message is very likely to map to a different ciphertext. ", "sentence2": "given two ciphertexts the adversary cannot distinguish whether they correspond to two encryptions of the same message or encryptions of two different messages.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2960", "sentence1": "This allows him to label columns of A with the corresponding attribute values.", "sentence2": "given prior information about a specific record, such as its location in a dataset or the value of its attribute, he can infer other attributes for that record (i.e., the pick up day for a taxi ride.)", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2961", "sentence1": "Though intuitively, batching makes it harder to extract precise information from the dataset, we show that it does not always suffice.", "sentence2": "if the data is ordered by some attribute, the information about a batch will provide information conditional on values of that attribute.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2962", "sentence1": "Instead of a secret key, the attacker can directly use the implementation itself as a larger effective key.", "sentence2": "he can isolate the program code where the key is embedded in order to copy the functionality of encryption/decryption routines and to utilize it in a stand-alone manner.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2963", "sentence1": "The function F is an T -secure weak white-box implementa­tion of EK if it is computationally hard to obtain an equiv­alent key of size less than T given full access to F .", "sentence2": "it should be computationally hard for an attacker to find any compact equivalent function which is smaller than T .", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2964", "sentence1": "Indeed, although more layers make the construction more secure, recent cryptanalysis [6] suggests that even as many as 9 layers (SASASASAS) are susceptible to attacks.", "sentence2": "the assurance on the security of (AS)i against decomposition is yet to be provided.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2965", "sentence1": "It seems unlikely that adversaries in practice will know all doc­uments.", "sentence2": "assuming knowledge of no documents is a step too far: the adversary may incidentally know that a user has one or more widely-circulated emails in her repository, for ex­ample.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2966", "sentence1": "This scheme supports not only keyword search queries, but also boolean and phrase searches.", "sentence2": "stemming, wildcard, and approximate-match searching are not supported.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2967", "sentence1": "Their attacks, and the more general statistical at­tacks alluded to by Song et al., are possible even when a scheme has been proven secure under a standard assumption.", "sentence2": "first we fix some notation from the SE literature.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2968", "sentence1": "This scheme differs from the above in that each row of the index is encrypted as a whole (say with a block cipher in some randomized chaining mode), so that no repeated document IDs can be read from the index before queries are issued.", "sentence2": "the length of each row is not hidden; before any queries are issued, the server can observe the result count for every row.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2969", "sentence1": "Note that both the ILP and Nomad Greedy optimize the total In­foLeakage.", "sentence2": "it is possible for Nomad Greedy to outper­form the ILP solution w.r.t.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2970", "sentence1": "Then, the only InfoLeakage client pairs affected are clients whose VMs reside in machine 1 and 2 (i.e., only the co-residency between VM c,i and VMs in machine 1 and 2 are affected by this move).", "sentence2": "we consider it inefficient to try the moves across all machines when the number of affected co-residency pair is bounded.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2971", "sentence1": "Lastly, Equation 16 models the migration budget such that the total number of migrations should not exceed PercentBudget of the total workloads (i.e., total number of VMs that currently have place assignments).", "sentence2": "d was scaled such that the number of epochs, hence the number of migrations, in each experiment is roughly the same for both exper­iments (i.e., D = 4 min for 20 min completion time and D = 1 min for 3 4 min completion time).", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2972", "sentence1": "Furthermore, with techniques like incremental diffs, the transfer size will be much less than the base VM memory image with 50% reduction (e.g., [18]).", "sentence2": "we expect minimal network impact from Nomad, especially given that datacenters handle much larger (tens of GB) flows [17].", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2973", "sentence1": "Unfortunately, solving this ILP is intractable and it takes more than a day to even solve a small problem instance with just 40 machines (Table 1 in §8).", "sentence2": "while the ILP approach is an exact optimal solution in terms of the leakage subject to fixed migration budget, it is far from viable in terms of the scalability requirements.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2974", "sentence1": "Recomputing the benefit (Line 6 in Algorithm 1) is a large contributor to high run time of the baseline greedy.", "sentence2": "we use an incremental benefit computa­tion which computes the delta in the current value of the objective function by only updating information leakage for set of dependent client pairs whose InfoLeakage are affected by the move.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2975", "sentence1": "That said, extending the basic algorithm to support fairness goals is an interesting direction for future work.", "sentence2": "defining suitable fairness objectives in a heterogeneous environment where different clients have varying number of VMs, degrees of replication, and sensitivity to migration is an interesting direction for future work.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2976", "sentence1": "Note that each move conceptually changes the state of the system and thus the benefit of future moves may decrease or increase de­pending on the moves we have already made; e.g., moving VM c,i may mean that all previously considered swaps involving this in­stance may no longer provide any value.", "sentence2": "nomad offers a practical vector-agnostic and robust defense against arbitrary (and future) side channels.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2977", "sentence1": "Fast side-channel attacks, in principle, could be addressed by reducing D accordingly (i.e., for side-channel attacks capable of extracting the key in 2 3 min, we suggest D = 30 sec).", "sentence2": "decreasing D comes at a cost of performance degradation.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2978", "sentence1": "This is in contrast to the Sum whose delta only depends on one input's value before and after an update.", "sentence2": "equation 9 in Figure 10 states that the total number of VMs as­signed to a machine should not exceed machine's capacity.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2979", "sentence1": "During the SSH handshake, the client and server select the client's highest priority mutually supported key exchange algorithm.", "sentence2": "we cannot directly measure what algo­rithm servers will prefer in practice.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2980", "sentence1": "At first, we thought that it might have been a complete API Update since the changes were so drastic.", "sentence2": "after computing the structural identity of x, the signature computation proceeds in the bottom-up fashion to compute signatures of the parent nodes.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2981", "sentence1": "Our whitelisting logic must guarantee that only the authorized scripts (scripts with legitimate signature) are allowed to enter the browser's JavaScript parser module, as mentioned in rule 1 in Section 3.2.", "sentence2": "our whitelisting module must be placed in a location where it is exposed to all the interfaces JavaScript parser has with client-side web components like HTML parser and CSS parser. In Chromium, the right point to place our whitelisting log.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2982", "sentence1": "Furthermore, SI C I L I AN can be fully applied to 372 domains with updates required only once in a month.", "sentence2": "sI C ILI A N covers five times more domains than SRI, which is based on raw signatures (69 domains).", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2983", "sentence1": "Sim­ply using type and initialization value as the identity might lead to semantically different code having the same structural signature.", "sentence2": "we propose to use the position set of the identifier in the statement to further refine its identity.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2984", "sentence1": "Such highly dynamic scripts include advertisements with customized scripts [28] or scripts from news websites.", "sentence2": "such \"non-static\" scripts are scarce (0.62%) with respect to all the crawled scripts and used only in 69 domains.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2985", "sentence1": "In our study, 97.99% of the scripts remain mostly static: no new code is introduced in these scripts.", "sentence2": "we believe whitelisting via signatures can be as effective in preventing script injection as it has been in securing desktop application in binary form.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2986", "sentence1": "For traditional cur­rency systems, any payment whether or not it is based on secure computation requires trust in a third party (as the currency it­self is based on a trusted party, such as a central bank).", "sentence2": "the introduction of cryptocurrencies, such as Bitcoin [5], opens the possibility of handling payments in a decentralized manner [6, 7].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2987", "sentence1": "Once all the community cards are revealed, parties that wish to claim the pot start revealing their cards.", "sentence2": "parties execute an additional stage where they take turns to reveal their cards, i.e., reveal their share hi,i (which reveals their hand).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2988", "sentence1": "Although we now handle a reactive functionality, we stress that the cash that is deposited at the beginning of the protocol is dis­tributed only at the end (i.e., no cash distribution occurs in any intermediate stage).", "sentence2": "secure cash distribution provides a means to keep the cash deposited in escrow while parties learn output from each stage's function evaluation, and thus can revise their inputs to a later stage.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2989", "sentence1": "When it's an honest party's turn to make a claim, it makes the claim if it possesses all the witnesses necessary for making the claim.", "sentence2": "an honest party may go ahead and claim a deposit even if (1) some deposits were not made, and (2) some claims were not made.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2990", "sentence1": "In practice, a ker­nel exploit may depend on system con.gurations, address space layout, compiler options, etc..", "sentence2": "to success­fully root a device, multiple exploits are usually attempted in both the malware [43] and the root providers.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2991", "sentence1": "Due to its privileged position, targeting Linux Kernel is natural to achieve full control over an Android device.", "sentence2": "a vulnerability in the kernel has a large impact as all devices that run the vulnerable kernel can potentially be affected.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_2992", "sentence1": "It is important to realize the PC-side program and independent Android app contain duplicate functionality of reading device information and retrieving exploits from local or remote store.", "sentence2": "as long as either one has a weak protection, the procedure can be revealed and exploits maliciously retrieved.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2993", "sentence1": "Note that we sort the table by the number of exploits we can find, yet it does not correspond to the same order presented in Table 1, therefore, not revealing the provider names.", "sentence2": "provider 6 and 7 only equip some basic protection in their ex­ploits, which is easy to bypass.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2994", "sentence1": "Vulnerable tamper-detection mechanism Signature or package name based tamper-detection can be found in many providers exploit files.", "sentence2": "the detection is executed only one time at the beginning, which makes it easy to by­pass modifying one conditional jump suffices and works in all cases.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_2995", "sentence1": "Theoretically, a kernel vulnerability affects all kernel versions between when the vulnerability is introduced and when it is fixed.", "sentence2": "simply counting the number of exploit binaries can be biased.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2996", "sentence1": "SafeDSA relies on a whole frame of the received samples to perform the cyclic prefix length estimation.", "sentence2": "we are curious about how many samples are good enough for the cyclic prefix length estimation.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2997", "sentence1": "The mobile switching center (MSC) in the circuit-switching domain transfers voice calls, and the 3G gateways enable data communication in the packet-switching domain.", "sentence2": "to call attention to this matter, this paper presents a systematic security analysis.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_2998", "sentence1": "The CP only allows legitimate call related requests and blocks all other packets utilizing the VoLTE interface from the AP.", "sentence2": "an adversary cannot send manipulated packets through the VoLTE interface.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_2999", "sentence1": "Furthermore, data encryption (e.g. IPsec or TLS for signaling, and sRTP for media data) should be deployed as specified in the 3GPP specifications [4, 5].", "sentence2": "even with strict binding and encryption, an adversary can still utilize tunneling since she has all permission for her phone.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3000", "sentence1": "In the case of Korea, video calls from operators cost 1.66 times the price of a voice call.", "sentence2": "we utilize RTP tunneling as all the operators we tested encapsulated voice data with RTP.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3001", "sentence1": "Despite this fast-moving trend, little research has been conducted to systematically examine security issues in the upcom­ing VoLTE services, not only in terms of their end-facing interfaces but also their cellular infrastructure.", "sentence2": "absence of authentication is one of the three threats originating from mis-implementation of the IMS network.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3002", "sentence1": "Therefore, cross-checking is a strong yet easy to implement solution.", "sentence2": "it cannot prevent caller spoofing in cases where an adversary spoofs the victim's IMS registration procedure, such that the SIP server stores spoofed parameters.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3003", "sentence1": "In our analysis, however, some operators do not follow the standard for call related services.", "sentence2": "this is another requirement for securing VoLTE.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3004", "sentence1": "We also found that when the call is initiated by the application, the calling state is not displayed on the phone.", "sentence2": "a user would not be able to recognize that her phone is now calling.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3005", "sentence1": "In this case, when a UE sends an INVITE message to a SIP server, the UE receives a response mes­sage with the other UE s IP address.", "sentence2": "a UE can collect another UE s IP address by randomly sending INVITE messages to the SIP server.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3006", "sentence1": "The throughput is adequate and there is still enough speed to transfer files as a torrent because there are many peers in the cellular network, and these peers do not usually shutdown their phones.", "sentence2": "people can share movies or other content when they are sleeping.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3007", "sentence1": "In contrast, full GPU virtualization has to emulate accesses to all GPU objects of the VMs scheduled to access the GPU.", "sentence2": "it has to emulate a wide variety of accesses to all GPU configuration registers7 and thus requires a large trusted code base.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3008", "sentence1": "However, once assigned to a SecApp, the GPU cannot be accessed by untrusted commodity OS/App code until the device is re-assigned to that code.", "sentence2": "a commodity OS/App cannot display its content during SecApp's exclusive use of the trusted display, unless the OS/App trusts SecApps unconditionally.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3009", "sentence1": "supporting native GPU drivers/Apps re­quires emulating all accesses to all GPU con.guration regis­ters for the VMs scheduled to access the GPU.", "sentence2": "adopt­ing full GPU virtualization for high-assurance trusted dis­play would be impractical.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3010", "sentence1": "Glider [41] could also be used to provide a trusted display service since it isolates GPU ob­jects in the security kernel.", "sentence2": "these approaches are unsuitable for unmodi.ed commodity OSes, because security kernels are object-code incompatible with native commod­ity OSes.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3011", "sentence1": "Our system solves the problem by sharing only the GPU display function between the untrusted OS/Apps and the SecApps.", "sentence2": "it needs to mediate only accesses to GPU objects that affect trusted display’s security.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3012", "sentence1": "Furthermore, only a small number of sensitive GPU objects require function emulation and this takes only a small amount of code, as shown in Section 5.1.", "sentence2": "implementing the GPU functions themselves (e.g., the large and complex native GPU drivers) within the GSK becomes unnecessary.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3013", "sentence1": "As GPU page tables support read-write access control in many modern GPUs (e.g. Nvidia GPUs [4], AMD GPUs [6] and recent Intel GPUs [23, 2]), the TDK can protect GPU command buffers by mapping their read-only accesses to GPU local page tables.", "sentence2": "some Intel GPUs provide different hardware protection mechanisms. ", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3014", "sentence1": "In our implementation, the fast channel takes 3.60 µs for the communication round-trip between mHV and TDK when they run on different cores.", "sentence2": "when the mHV needs to switch to TDK and back on the same core, the overhead is 722.42 µs.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3015", "sentence1": "For example, instruc­tions are executed on GPU processor cores, process input data, and produce results that are used by display engines.", "sentence2": "commands are executed by dedicated command processors and are used to configure the GPU with correct parameters; e.g., specify stack base address used by instruc­tions.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3016", "sentence1": "In­tuitively, the security-sensitive GPU objects are those that can be programmed by untrusted software (e.g., malicious drivers, applications) to break the confidentiality or authen­ticity of trusted display output, and those which can be tainted by access to other sensitive GPU objects.", "sentence2": "new hardware architectures make the trapper's overhead negligible.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3017", "sentence1": "The TDK has to mediate access to far fewer GPU configuration registers than full GPU virtualization.", "sentence2": "access to 39 out of 625 GPU configuration registers require mediation, 13 of which are needed for hardware overlays.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3018", "sentence1": "Once a user has created an account for such a username-password combination with the LS, he can subsequently login by providing the correct username and pass­word again.", "sentence2": "the login server must be able to verify whether a password attempt pwd ' matches the stored pass­word pwd or not.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3019", "sentence1": "Just recently, a novel probably approximately correct (PAC) learning framework has been developed, which ensures the ultimate delivery of a mathematical clone for pre-defined levels of confidence and accuracy [10].", "sentence2": "arbiter PUF manufacturers have been forced to propose countermeasures including XOR arbiter PUFs as an effective technical action [34].", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3020", "sentence1": "To explain our lattice basis reduction attack in an easier manner, we will assume a slightly different situation. ", "sentence2": "we emphasize that for the real distribution a more complex analysis can be performed that would yield an (asymptotically) equivalent result.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3021", "sentence1": "The total number of disclosed coefficients is counted separately in order to avoid having common disclosed vectors in the three centering groups.", "sentence2": "we have observed that no common disclosed vector is found in the three groups.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3022", "sentence1": "We launched our attack on simulated CRPs, generated in Magma.", "sentence2": "it is worth noting that our XOR arbiter PUF simulator implemented in Magma adopts only real ex­perimental data obtained in [35].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3023", "sentence1": "Attack attribution to identify attackers is generally considered as a hard problem because attackers may employ various methods to conceal their identities, e.g., obfuscating or re-packing binaries, changing the URLs and domains of servers, launching attacks from geographicallydistributed compromised machines.", "sentence2": "55.5% of malicious downloaders are signed, accounting for 73.8% of the malicious influence graphs.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3024", "sentence1": " However, node B downloads additional files, some of which (nodes C and F) are detected as malicious, suggesting that node B, and potentially node D downloaded from the same domain as node B, as well as all of the nodes reachable from them in the downloader graph, are likely involved in malware distribution.", "sentence2": "by analyzing the downloader graphs on a host we can identify large parts of the malware download activity, which may otherwise remain undetected.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3025", "sentence1": "The protocol con­sists of two rounds.", "sentence2": "each trustee Ti performs the following: Round 1: 5Note that, during this step, any invalid ciphertexts and duplicated ciphertexts are removed.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3026", "sentence1": "Finally, A posts the election transcript info to the BB..", "sentence2": "the adversary can still copy and re-randomize some honest voters ciphertexts as well as their attached NIZK proofs if the same CRS is used among all the voters.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3027", "sentence1": "It is easy to see that as long as the master CRS is of the perfectly sound type, the proofs that crs0/1 is an encryption of 1 will be perfect and hence the voters cannot stuff invalid ciphertexts in the virtual ballot box.", "sentence2": "as mentioned above, the EA might attempt to use a master CRS of the second type (simulatable) and then collaborate with a voter to violate integrity.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3028", "sentence1": "Our simulator is given as input the number of attackers to generate, as a percentage of population size.", "sentence2": "as the population increases we assume a linear increase in at­tackers as well.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3029", "sentence1": "That information can be used for sorting/filtering a student list without additional metadata.", "sentence2": "student ID will change when the student moves to another department or proceeds from undergraduate to graduate school, so he/she loses personal data when Student ID changes.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3030", "sentence1": "The manual surf exchanges enforce that one account be used only from one IP address at a time.", "sentence2": "all the autosurf exchanges we studied allow an account to be used from multiple IP addresses simultaneously.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3031", "sentence1": "By contrast, Table 3 shows that traffic from the manual exchanges is predominantly directed at sites that advertise affiliate programs, other manual traffic exchanges, or pages advertising goods sold by the participants themselves.", "sentence2": "pages visited in the manual exchanges appear directed at luring humans (the exchange participants) to upgrade, join other exchanges or affiliate programs, or purchase products that do not have mainstream appeal.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3032", "sentence1": "Javascript on this page automatically fetches the next site to be viewed periodically and opens it inside an iframe within the surf page.", "sentence2": "we calculate that dispersed clicks can be produced at about $0.11 per thousand.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3033", "sentence1": "Table 7 presents a detailed comparison of the coverage of our seeded method to the coverage of the baseline method.", "sentence2": "comparing columns four and six shows the total number of domainkeys selected by the two methods in each category, respectively.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3034", "sentence1": "In the case of email addresses, for example, 21% of domain-keys have all their values matching the rule (i.e., the ratio is 1) and over half (58%) of the domain-keys have at least 20% of their values matching the seed rule.", "sentence2": "the name domain-key shows that the vast majority (over 90%) of domain-keys that have at least one matching value have less than 10% of all their values match.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3035", "sentence1": "The intuition behind this is that key names will likely be used coherently (i.e., for carrying the same type of information) within the same domain (e.g., google.com may use the keyword ggender to collect user s gender re­gardless the specific Google service).", "sentence2": "a same key might be used in the context of different domains with a different meanings (e.g., the key id may be used dif- Dataset HTTP flows Tuples Domain-keys Lab traffic 9,227 20,810 8,372 ISP traffic 40,775,119 51,368,712 3,113,696  Table 2: High-level statistics of our two datasets.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3036", "sentence1": "The resulting infrastructure makes these services conveniently available anytime and anywhere, enabling them to become an integral part of daily life. ", "sentence2": "users explicitly and implicitly provide a wealth of Personal Information (PI) that reflects several aspects of their life.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3037", "sentence1": "For this dataset, we wanted to be able to examine the TLS/SSL encrypted traffic as well.", "sentence2": "we collect data by tapping in to the connections with a middlebox that passes traffic through VPN/Man-In-The-Middle (MITM) transparent setup [15].", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3038", "sentence1": "Table 1 contains the number of instructions for each function needed to create a new user.", "sentence2": "an adversary with knowledge that the scheme has been used will be able to retrieve HDF(passwd username) by XOR ing the salt with the ersatzpassword.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3039", "sentence1": "One way to achieve that to check the binary of the authentication modules.", "sentence2": "this will only inform an adversary that these modules have been changed without giving any details of how they have been changed unless they can be exfiltrated and disassembled.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3040", "sentence1": "Security administrators can easily configure their systems to forward adversaries to a fake account where they can be further monitored and observed. ", "sentence2": "our scheme comes with an intrinsic incentive that security administrators will not only enhance the security of password storage, but also detect when such files are leaked and cracked.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3041", "sentence1": "This property benefits not only the early adopters of the scheme, but also the overall security of other (non-adopting) systems.", "sentence2": "in this case, an adversary has access to the system, where she can observe authentication attempts, has access to user­names/passwords files, and can query the hardware-dependent function HDF; the only thing an adversary does not have ac­cess to is plaintext passwords and the secret HSM key.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3042", "sentence1": "It would also look sus­picious if all user passwords were crackable.", "sentence2": "it would be wise to use some randomly-generated ersatzpasswords within a system to enhance the scheme's plausibility.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3043", "sentence1": "The Pin-based version of DynaGuard incurs an average overhead of 170.66%.", "sentence2": "this overhead is dominated by the native DBI framework (Pin), with DynaGuard adding only 2.92% on top of that, on average.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3044", "sentence1": "The canary initialization that occurs during the creation of threads and processes is exactly the same in DynaGuard and in glibc, with the only difference being that the Dy­naGuard canary is stored at a different location in the TLS area.", "sentence2": "the entropy of canaries is not affected, but now the TLS holds two different types of canaries: the stan­dard glibc canary and the DynaGuard canary.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3045", "sentence1": "However, this is not feasible in current canary-based pro­tectors, as hardened programs do not store any information regarding where (in their address space) their canaries are stored.", "sentence2": "once a child process is forked, there is no way for it to access the canaries in the frames it inherited from its parent process, as it cannot differentiate canaries from random data that may be residing in the stack.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3046", "sentence1": "Thus the increased resilience to shoulder-surfing attacks comes at the price of additional storage and computation time.", "sentence2": "with respect to the verification operation, the increased computation time can also be seen as an advantage.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3047", "sentence1": "The purpose of this study is to compress the images while ensuring their security.", "sentence2": "we propose an ETC technique through which we can achieve good compression performance while ensuring the security of images.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3048", "sentence1": "Hence, as discussed above in Mishra et al [12] scheme any legal user can achieve the identity of any other legal user.", "sentence2": "we can confirm that Mishra et al scheme fails to achieve user anonymity.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3049", "sentence1": "MUDFLOW is not the first work that uses machine learning techniques to detect malicious Android applications.", "sentence2": "it is the only one, together with CHABADA, to train the model only on benign applications [12].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3050", "sentence1": "We use the data-flow differences to automatically identify malware based on its flow of sensitive data (Section V).", "sentence2": "we train a model using data flows of benign apps only, and use it to detect novel malware even if no earlier malware samples are known.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3051", "sentence1": "If the resulting value is greater than a certain threshold, namely the two C-SI matrix is similar, indicates the user is currently in a static scene.", "sentence2": "the user is in a dynamic scene.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3052", "sentence1": "If there exists an attack (malicious client) who continuously reports forged CSI, the whole networks will be completely break down for UL MU-MIMO transmission.", "sentence2": "this solution may slightly reduce the effective time of CSI, and drop the data transmission efficiency down.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3053", "sentence1": "Considering the nodes in office LAN often is stationary and the channel between clients and AP changed slowly, meanwhile the nodes in mobile LAN often is dynamic and the channel between clients and AP changed quickly.", "sentence2": "the CSI's period of validity in stationary scenarios is longer than in dynamic scenarios.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3054", "sentence1": "In our view, the accuracy of CSI is more important than SNR even though it is useful in the process of decoding in MU-MIMO system.", "sentence2": "cSI measurements are often subject to errors, induced by unreliable hardware, antenna connectors or even the hostilely forged CSI.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3055", "sentence1": "The idea is to perform self-benchmarking of critical portions of code in order to ensure that debugging breakpoints have not been inserted, and an attacker has not tampered with critical portions of the virtual processor such as the native system calls, using timing analysis from within the V-OS firmware to detect such an attack in progress.", "sentence2": "while this concept works against some attacks, we have found this difficult to implement in production due to the native processor speed variation both between mobile devices as well as within an execution session due to multitasking.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3056", "sentence1": "Because the links between the application and its libraries are only resolved at load time and because they are stored in the form of easily readable symbol information, attackers can study which external functionality the program requires and influence the linking process to their own advantage.", "sentence2": "attackers can hook a shared library, which means that run-time invocations to shared library functionality are redirected to code controlled by the attacker.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3057", "sentence1": "However, in a statically linked program, the program loader no longer performs any actions on behalf of the code that has been statically linked into the program.", "sentence2": "our prototype then links those libraries into the application, after which instructions that indirectly refer to library symbols through the IAT are rewritten to refer directly to the now linked-in symbols instead.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3058", "sentence1": "To prevent attackers from easily reverse engineering the distributed binaries, a software vendor can obfuscate the binaries he distributes, i.e., transform them into semantically equivalent ones that are significantly harder to understand and reverse engineer.", "sentence2": "when a vendor deploys an obfuscator tool on his software, this tool will only transform the code available at obfuscation time.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3059", "sentence1": "Thirdly, the details of the internal state kept by ntdll differ across versions.", "sentence2": "the state initialized by the ntdll automatically loaded by Windows can differ from how the ntdll statically linked into the program expects it.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3060", "sentence1": "Custom Library Loader The original program entry point (point 1 in Figure 2 (a)) is replaced with a newly inserted library loader function (point 1 in Figure 2 (b)) that replaces the operating system s dynamic linker.", "sentence2": "attackers can hook a shared library, which means that run-time invocations to shared library functionality are redirected to code controlled by the attacker.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3061", "sentence1": "In spite of the fact that it is very important to evaluate the stealth of the protected code, the method for evaluat­ing the code stealth is sorely lacking.", "sentence2": "in this paper, we propose a metric for measuring code stealth.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3062", "sentence1": "The results of this case study show that optimization and static obfuscating transformations (e.g., replacing with fun­damental instructions, embedding a specialized interpreter, flattening control flow and encoding arithmetic/data) have little effect on artificiality.", "sentence2": "dynamic obfuscating transfor­mations (e.g., instruction camouflage and code encryption), or a technique that inserts junk code fragments into the program, tend to increase the artificiality, which may have a significant impact on the stealth of the code.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3063", "sentence1": "In a simplified view, a license check can be seen as a predicate p, which either evaluates to true or false.", "sentence2": "the output of the SMT solver that satisfy the constraints leading to the true branch of the license check represent a concrete license key value for the attacker.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3064", "sentence1": "Nowadays, the XACML (eXtensible Access Control Markup Language) [1] has become the de facto standard in the context of access control systems.", "sentence2": "the second row reports the name of mutation operators, the first column the Sun PDP classes involved in the experiment while in each table entry there is the obtained mutation score (the dash indicates that the mutation operator is not applicable to the Sun PDP class).", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3065", "sentence1": "It is necessary for the Matelas model to be accessible and understood by such platforms.", "sentence2": "the Matelas model is translated into the Java Modelling Language (JML) [9].", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3066", "sentence1": "In the first one, a framework-aware application is used, e.g. one that knows how to invoke the functionalities of the enforcement & publishing layer, which are exported through a set of layer API.", "sentence2": "coco Cloud supports several deployment and service models, since its services may be offered in a public, private, or hybrid Cloud, and in IaaS or SaaS services.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3067", "sentence1": "Outsourcing data storage and management to the Cloud, usually for improved availability and sharing, entails an inherent loss of control by Cloud tenants.", "sentence2": "additional means of data protection and usage control must be provided to allow organizations to safely share their data on the Cloud.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3068", "sentence1": "At run time, the value of each field that was labeled as a source of private data is logged.", "sentence2": "the most basic extensions execute standard class behavior, but have code that allows Labyrinth to log all the private data entered by end users, such as user ID, password, and credit-card number, as well as data that the application retrieves from the environment, such as the device s location.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3069", "sentence1": "To guarantee authentication and integrity of the protocol, R-CARP employs digital signatures. ", "sentence2": " we have selected the Boneh-Lynn-Shacham [9] (BLS) scheme, which produces short signatures of 160 bit providing a security level of 280.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3070", "sentence1": "There were also experiments [8, 11, 12, 14] carried out on character-based passwords.", "sentence2": "as the scope of our work in this paper is on numerical PIN inputs, we will not discuss character-based experiments any further.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3071", "sentence1": "To reflect real world situations as much as possible, the demography of the subjects taking part in data collection should be as diverse as possible.", "sentence2": "people from different age groups, of different genders and with different device usage frequencies (this indirectly correlates with device familiarity) should be represented as much as possible.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3072", "sentence1": "The garbled RAM construction and security proof above can be generalized to a setting in which the garbled RAM programs act on a persistent database.", "sentence2": "the updates that a garbled RAM program makes to a database D are accessible to the next garbled program to be executed on that database.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3073", "sentence1": "As discussed in the introduction, in applications where the memory access pattern is known not to leak sensitive information, this notion of garbling may be significantly more efficient.", "sentence2": "it preserves the efficacy of cache, for which real-world RAM programs are extensively '' ''' optimized.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3074", "sentence1": "Bitansky Garg et al. and Canetti Holmgren et al. con­struct a semi-succinct garbling scheme for RAM programs, assuming non-succinct Indistinguishability Obfuscation (IO) and injective one way functions [BGL+15, CHJV15].", "sentence2": "they construct garbling schemes where the space and run­time of the garbled program are proportional to the space and runtime of the plaintext program, and where the size of the garbled program is proportional to the space complexity of the plaintext program.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3075", "sentence1": "When the plaintext program is a Turing ma­chine, making sure that the memory access pattern is fixed incurs only small overhead in complexity.", "sentence2": "hid­ing the memory access pattern in a RAM program in an efficiency-preserving way requires the memory access pat­tern to be randomized.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3076", "sentence1": "We construct the first fully succinct garbling scheme for RAM programs, assuming the existence of indistinguisha­bility obfuscation for circuits and one-way functions.", "sentence2": "the size, space requirements, and runtime of the garbled program are the same as those of the input program, up to poly-logarithmic factors and a polynomial in the security parameter.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3077", "sentence1": "Koppula, Lewko and Waters [KLW15] devise a fully suc­cinct garbling scheme for Turing machines from non-succinct IO and one way functions, using techniques that extend those of [CHJV15].", "sentence2": "in their garbling scheme the runtime, space and description size of the garbled program are proportional to those of the Turing machine representa­tion of the plaintext program.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3078", "sentence1": "The client would want both storage and computation required for secure delegation to be significantly less than the actual data and computation.", "sentence2": "the server, who is actually storing the data and performing the computation, would like to operate on the (private) data as a PRAM program, and not to perform too much work when compared to computation on the plaintext data.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3079", "sentence1": "Note that a memory block is the smallest unit of opera­tion in CP ORAM, which consists of a fixed small number of memory cells.", "sentence2": "a position map pos (stored in CPU state) records where each memory block is stored in the tree, i.e., a node somewhere along a path from the root to the leaf indexed by pos[g].", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3080", "sentence1": "To handle the full-fledged PRAM computa­tion, we introduce several techniques to tackle the challenges in the parallel setting.", "sentence2": "we avoid the depen­dency on the number of CPUs to achieve full succinctness.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3081", "sentence1": "So, we can view these pj's as public states of OPAccess, and do not erase its content in the hybrids.", "sentence2": "we generate simulated path pj for each CPU, and store them as public states to simulate the access pattern of OPAccess.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3082", "sentence1": "Note that normal signatures and Merkle tree cannot guarantee functional equivalence as forgeries ex­ist information-theoretically.", "sentence2": "more importantly, we show how to use our (fully suc­cinct) CiO-PRAM to construct the .rst fully succinct ran­domized encoding (RE) for PRAM computation.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3083", "sentence1": "To argue for the indistinguishability between Hybi and Hyb ' i, we note that the i-th decryption key is not used in the honest evaluation, since the computation after time step i is erased.", "sentence2": "we can puncture the random­ ness and erase the decryption key from the program (which uses CiO security as well), and reach a hybrid (from Hybi) where semantic security holds for the output ciphertext at time step i.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3084", "sentence1": "Toward that end, we propose a scale development study to specify the components of H and I in Figure 1 and subsequently test the entire model (i.e., letters A through I in Figure 1).", "sentence2": "we propose a study aimed at specifying perception-based trust in a way that is quite novel because it is grounded in the trust requirements and perceptions of system administrators.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3085", "sentence1": "We will operationalize important trust items as items with high factor loadings on factors with large eigenvalues.3 What will not be immediately clear is exactly which items are the strongest indicators of trust.", "sentence2": "for these reasons, we will analyze our data focusing on items with high factor loadings on factors with large eigenvalues.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3086", "sentence1": "For the Sancus-based design of Soteria we thought of two possible concepts: (1) Putting a loader stub together with the code that should be encrypted into one module and decrypting it in-place, and (2) designing a separate loader module.", "sentence2": "we give upper bounds for the actual performance results which are expected to be even less than those shown in Table 2.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3087", "sentence1": "Currently, many electronic door access control systems use RFID smartcard as part of its authentication process [4][5][6][7].", "sentence2": "for normal RFID smartcard process, when the UID is available in the database system, the user is granted authorisation to access the door and this can be easily breached.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3088", "sentence1": "LPP is obtained by finding the optimal linear approximations to the organ functions of the Laplace Betrami operator of the manifold [21] [22].", "sentence2": "lPP is a linear technique to recover important aspects of the intrinsic nonlinear manifold structure by preserving local structure.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3089", "sentence1": "Since, if it happens it can cause loss of properties, traumatized, and most worst is can threat human life.", "sentence2": "the development of this system is significant where it is able to prevent from the fire hazard.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3090", "sentence1": "This sensor is not only sensitive to smoke, but also to flammable gas [6].", "sentence2": "it is suitable to implement at home for surveillance security.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3091", "sentence1": "In this implementation, it can be used only for unlocking screen of personal computers since the accuracy of personal identification is not good.", "sentence2": "if signs are managed as a part of password like the method by Chahar et al., the accuracy may extremely increase.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3092", "sentence1": "In their method, individuals are identified by four types of verification with a multi-touch panel.", "sentence2": "phase Registration of information is required before users use the system.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3093", "sentence1": "The sign to be registered can be not only a sequence of characters but also everything e.g. an illustration since just motion of fingers are used for identification.", "sentence2": "complex images take long time to be drawn and the details of complex images are hard to be drawn with the same way in every time.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3094", "sentence1": "In their research, examinees write a mark of plus in the air.", "sentence2": "the software can be used only for unlocking the screen of a smart phone.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3095", "sentence1": "To achieve this, DRM-protected files are securely encrypted with a cryptographic key, which can only be decrypted by authorized DRM agents under specific usage rules and condi­tions.", "sentence2": "the effectiveness of these DRM technologies in prac­tice is still controversial.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3096", "sentence1": "The main motivation of our experiments is to analyze potential risks of OMA DRM services and to suggest reasonable countermeasures to mitigate such risks.", "sentence2": "we evaluated whether DRM protection can be disabled through a case study of MelOn.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3097", "sentence1": "Also, the DRM content can be re-distributed in a local network through a distribution mechanism called super-distribution.", "sentence2": "the roles of these components are very important since the overall security of a DRM system depends on the secure implementation of a DRM agent and its relationship to the other components.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3098", "sentence1": "The jnz instruction is two bytes long.", "sentence2": "we should re­place this instruction with other instructions that are also two bytes long in order to keep the other instructions at the same positions in the original code.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3099", "sentence1": "In this paper, we demonstrate how to bypass the integrity checking of the rights object in the OMA DRM system through a case study of MelOn (a well-known music distribution service in South Korea) by reverse engineering its media player equipped with a DRM agent.", "sentence2": "we need to disable the integrity check of the rights object in the OMA DCF files if we want to use it without any usage constraint.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3100", "sentence1": "As we already described in Section 2.1, however, the integrity of the rights object in a DCF file is securely protected by its MAC.", "sentence2": "the DRM agent, embedded in the MelOn music player, always checks if any of the permissions and constraints included in a DCF have been maliciously modi.ed before playing the DCF file.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3101", "sentence1": "With a user accessed domain, as the dashed graph, the DSG graph shows a large flux variation, higher AWS, with many folding point.", "sentence2": "the DSG graph of an auto-ware accessed domain has just shown a little variation, lower AWS, as the solid graph.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3102", "sentence1": "Group having smallest centroid is auto-ware accessed domain, in other side, group having biggest centroid is user accessed domain. ", "sentence2": "the DSG graph of an auto-ware accessed domain has just shown a little variation, lower AWS, as the solid graph.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3103", "sentence1": "Finally, HTTP polling activity is pointed out by computing the standard deviation of the cluster.", "sentence2": "normal auto-ware (e.g. updater and downloader) transmits a similar periodic pattern of traffic that has been generated within a short period of time.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3104", "sentence1": "The remaining producers are honest-butcurious.", "sentence2": "these producers faithfully follow their protocols, but may try to exploit additional information that can be learned in doing so.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3105", "sentence1": "Federated identity services have gained widespread popularity among users as a simplifying means for managing their online identities.", "sentence2": "federated identity providers (e.g., Facebook and PayPal) employ authentication protocols such as OAuth [20, 21] and OpenID [34] to offer their users the convenience of using single identity and thus a single credential  to log into and access various third-party applications (e.g., Wiki and StackOverflow).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3106", "sentence1": "Penetration testing, and security assessments in general, are critical for any company that relies on an IT infrastructure.", "sentence2": "a penetration test procedure can take several weeks or months, depending on the size and complexity of the network being targeted and the level of the details that the client wants.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3107", "sentence1": "This system relies on a miniaturized computer to host the necessary components to conduct a security assessment.", "sentence2": "in another scenario, for a company of several thousand users and with a great variance in their infrastructure, the penetration testing process could take several weeks or even months to complete due to the complexity of the network/application architectures and many different attack avenues available.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3108", "sentence1": " For example, Google has developed most of its major products using this set of technologies (Gmail, Google Groups, Google Maps etc.).", "sentence2": "if a browsers leaves the system, the remaining actions will simply be distributed among the remaining browsers.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3109", "sentence1": "When our tool is able to guess immediately what user action was performed next, having several browsers working on the problem is not helping (although it doesn t hurt either).", "sentence2": "when finding the next action is problem­atic (which is the time-consuming scenario that is presented in Figure 4), D-ForenRIA fully benefits from having several browsers working in parallel.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3110", "sentence1": "DECAF relies on VMI to retrieve the running processes and loaded modules inside a virtual machine to analyze the behaviors of specified processes or kernel modules, for automatic malware detection and analysis.", "sentence2": "by adopting the strategy one in discussed in Section 4.3, the correct offsets can still be found by filtering the false offset values from the false 10 ORIs.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3111", "sentence1": "With the base register identified, flow-insensitive forwarddata-flow analysis on the base register reveals all the ORIs present in the function.", "sentence2": "in Figure 4, an ORI source at 0x402 is first identified via dynamic analysis.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3112", "sentence1": "In Proceedings of the 2010 workshop on New security paradigms, pages 1 6.", "sentence2": "in some cases defenders may also become active defenders by taking proactive measures to subvert the attacker's control while simultaneously executing their own mission objectives.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3113", "sentence1": "The time bound of the permutation-based overwriting comes from the performance characteristics of rotational drives, which are optimized for sequential file accesses but performs relatively poorly for random accesses.", "sentence2": "different from the previous works, KMF-based methods do not involve any third party to manage the data keys.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3114", "sentence1": "In this experiment, we con­sider a honestly cloud and a malicious cloud.", "sentence2": "if a malicious cloud performs the deletion operations on the fly without extra storage, it is practical infeasible for the cloud to pass the challenge-response protocol.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3115", "sentence1": "Logically, n500 does not affect scan_enable's function but the presence of an extra logic gate and a signal is unusual and needs to be examined by the user.", "sentence2": "a report of suspicious internal signals is generated.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3116", "sentence1": "Al­though [2] overcomes the major drawbacks of the recom­bined fingerprinting approach, it still exhibits the following shortcomings: (1) the system still requires a two-layer anti­collusion encoding of the fingerprint, which results in long fingerprints; and (2) the hash of the fingerprint is used to trace an illegal re-distributor in case of collusion of several buyers and, hence, such a hash must be a valid codeword of some anti-collusion code.", "sentence2": "the transaction monitor still requires the cooperation of the proxies to construct a valid hash-level codeword.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3117", "sentence1": "Software-based attestation [22, 28, 34, 45, 46] does not require any hardware and involves no cryptographic keys.", "sentence2": "other similar attack scenarios are possible, where the com­mon feature is the ability of malware (which is resident on compromised devices) to manipulate their clocks.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3118", "sentence1": "Such attacks start with full or partial de-capsulation (i.e., removal of packaging using mechanical or chemical means, or mixture of both), followed by de-processing.", "sentence2": "static network techniques can not be extended to dynamic networks.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3119", "sentence1": "To achieve this, we need to make some minimal assumptions about hardware security features of the underlying devices.", "sentence2": "the integrity measurement mechanism (attestation code) residing on Di must be immune to software attacks.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3120", "sentence1": "The second issue has not been considered at all in the context of remote attestation.", "sentence2": "the first issue has been noticed and progress has been made, to some extent, by Asokan et al. [4] in the design of SEDA– a technique for efficient and scalable attestation of a network of provers.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3121", "sentence1": "Android's permission system empowers informed privacy decisions when installing third-party applications.", "sentence2": "examining the access permissions is not enough to assess privacy exposure; even seemingly harmless applications can severely expose user data.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3122", "sentence1": "Viber keeps a configuration file with information on the user identity, profile picture and phone number.", "sentence2": "inside the Viber directory, the hidden file .userdata contains the username specified when registering with the messaging service.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3123", "sentence1": "As we can see, it is relatively easy to detect the user gender from user photos.", "sentence2": "the gender detection algorithm correctly identi.ed all photos of female users and misclassified only one male.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3124", "sentence1": "SEDA is based on neighbors verification and hop-by-hop MACs for authentication.", "sentence2": "every device in SEDA is supposed to be equipped with the minimal hardware required for attestation.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3125", "sentence1": "The se­curity of a password for example generally increases with its length, but so does the effort required to recall and enter it.", "sentence2": "a trade-off exists between security and cost which depends on the particular risk to defend against.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3126", "sentence1": " Writing in a noisy environment can significantly increase the difficulty of eavesdropping. ", "sentence2": "it is still possible to segregate the sound of handwriting by using multiple microphones [26]. ", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3127", "sentence1": "The features of sound fragments are calculat­ed by Java.", "sentence2": "before the test they were not aware that their data are used for testing an eavesdropping system.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3128", "sentence1": "However, this increases the space of candidate words for word recognizer and thus would reduce the accuracy even if the word has hit the database.", "sentence2": "if the database is too small, it is highly probable that the word written by the victim does not exist in the database, which leads to the failure of recognition directly.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3129", "sentence1": "A naive method is setting a constant threshold and once the sound signal magnitude or instantaneous power exceeds the threshold, a stroke is detected.", "sentence2": "since noise levels are usually different in different environments, and the aver­age noise power varies over time, it is hard to assign such a threshold.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3130", "sentence1": "Koppel et al. [31] identify 31 unique causes where healthcare professionals use workarounds to BCMA processes that they consider impractical (e.g., time).", "sentence2": "these workarounds can result in the wrong administration of medication which impacts patient safety.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3131", "sentence1": "Finally, we will discuss the open research problems about the security and privacy of genomic data.", "sentence2": "we will discuss (i) genomic data sharing between different enti­ties, (ii) privacy vs. utility of genomic data, (iii) integrat­ing genomic data in individuals electronic health records, (iv) access control issues, (v) credibility and liability issues of genomic data, and (vi) standardization efforts by Global Alliance for Genomics and Health (GA4GH).", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3132", "sentence1": "Such wide usage of genomic data also raises serious privacy-related concerns on the data.", "sentence2": "we will discuss some well-known and important attacks on privacy of genomic data.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3133", "sentence1": "To address zero day attacks, others approaches are needed.", "sentence2": "knowledge about common behavior of malware may be needed such as dynamic behavior of the attack.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3134", "sentence1": "We remark that KEA is non-falsifiable (cf. [42]), and it is indeed a strong assumption.", "sentence2": "one can argue that non-falsifiability might be inherent for extractable hash functions, and thus L-more extractability.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3135", "sentence1": "Each access control server is responsible for verifying the access control requirements bound to specific object/permission combinations, issuing signed tokens attesting to such verification, and verifying the signatures of the tokens it receives.", "sentence2": "each AC server must faithfully store a private token signing key in a secure manner and verify both the access control requirements governing specific token requests and the signatures and claims on all incoming tokens.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3136", "sentence1": "It is encompassing because an edge is defined to be clean only in case all of the files are clean.", "sentence2": "even one malware file out of a thousand clean files, for in­stance, classifies an edge as malware, and thus, the edge would be added to Gm t .", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3137", "sentence1": "Unfortunately, neither the length nor the content of the padding are checked for integrity during decryption.", "sentence2": "we cannot exclude that an attacker which sits on the same client and shares resources (such as memory, CPU, etc.)", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3138", "sentence1": "This is achieved by exploiting the ad­dress book syncing feature in OTT messaging applications.", "sentence2": "in addition, social information about the caller can be displayed, like number of mutual friends, presence on social networks etc.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3139", "sentence1": "Instructing them to focus on the toolbars in the browser did not make much difference since participants reported assessing the legitimacy of the website by how it looked and felt.", "sentence2": "the locus of attention and strongest cue was the page itself, not the chrome surround­ing it.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3140", "sentence1": "For privacy, the situation is rather different: even an accurate privacy warning may still be dis­missible depending on the user s specific privacy preferences.", "sentence2": "security is a yes or no, privacy is a spectrum.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3141", "sentence1": "The reason is that the (absolute) ai values obtained from SVM learning are bounded by the corresponding value of the SVM parameter C.", "sentence2": "if we set a lower C value for the malicious samples, their ai values will decrease.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3142", "sentence1": "In the case of evasion attacks, only malicious data at test time is affected.", "sentence2": "despite these techniques have been shown to be effective in some adversarial learning tasks, their adoption in practice is hindered by different factors, including the difficulty of meeting specific theoretical requirements, the complexity of implementation, and scalability issues, in terms of computational time and space required during training.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3143", "sentence1": "A query sent to a DNS revolver is resolved from the root of the hierarchy going down to the delegation system involving several servers in the world.", "sentence2": "introducing DNSSEC impacts several areas of the operation of an Authoritative DNS server [18].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3144", "sentence1": "The tickets are stored in the Request Manager linked to a particular request.", "sentence2": "results Collector's handlers can work on each particular request by using this ticket­request association.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3145", "sentence1": "We claim that implementation diversity is a reasonable way to justify the standard adversarial model in threshold cryptography.", "sentence2": "our solution provides by now implementations for the nodes in Java and Phyton, however the presented modular design allows to implement nodes in others programming languages.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3146", "sentence1": "The experiment results show that the algorithm is strongly robust against attacks from JPEG compression, cropping, noise and so on.", "sentence2": "as choosing the pixels in the least important point, the watermark information is destroyed easily by filter, image quantization, and geometric distortion.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3147", "sentence1": "In our study, we test both compare-and-confirm and compare-and-select.", "sentence2": "we explore compare-and-select due to its potential benefits for inattentive users.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3148", "sentence1": "This means, our fingerprints are being saved somewhere by the mobile phone operators.", "sentence2": "support for the Biometric Registration System A total of 36 people (20.9% of comments in the open-ended box) said that they supported the biometric registration program.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3149", "sentence1": "Our data shows that the situated idea of privacy among participants often made them resist the biometric registration program.", "sentence2": "the origin, nature, and characteristics of privacy in the Bangladeshi context has not been studied enough to explicate this resistance.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3150", "sentence1": "With the rapid growth of ICT adoption in the Global South, crimes over and through digital technologies have also increased.", "sentence2": "governments have begun to undertake a variety of different surveillance programs, which in turn provoke questions regarding citizens’ privacy rights. ", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3151", "sentence1": "One participant described: Do you think police do not know who the criminals are in a neighborhood?", "sentence2": "when the devices leave these stable environments, a whole new set of designs and policies are required to understand privacy in different scenarios.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3152", "sentence1": "Several studies on technology and gender in the Global South have shown how the relationship between technology and women is affected by the male-dominated cultural norms [3,5,36].", "sentence2": "we know little about their impact on the notion of privacy and implications for biometric identification.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3153", "sentence1": "Many participants were concerned about where their fingerprint data would be stored and how it might be used in the future.", "sentence2": "this deadline was subsequently extended for one month since, on the day of the deadline, the majority of SIM cards were still not registered [65].", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3154", "sentence1": "Unable to register these customers, the registration person suggested that they go and talk to their Ward Commissioners to obtain new IDs.", "sentence2": "several people reported that they had already talked to their Ward Commissioners but had failed to get new ID cards since they were not originally registered in their current Wards.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3155", "sentence1": "For example, the registration system required that the owner of a SIM card identify themselves with a valid ID, which could be their national ID card or passport.", "sentence2": "another said that he burnt his hand working with hot oil.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3156", "sentence1": "Similarly, different citizen groups monitor and criticize government surveillance programs [45].", "sentence2": "many countries in the Global South are struggling to maintain a stable democracy and are embarking on mass surveillance programs with little external oversight.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3157", "sentence1": "Despite this upgrade, we demonstrate that it is possible to accurately identify Netflix videos from passive traffic capture in real-time with very limited hardware requirements.", "sentence2": "we developed a system that can report the Netflix video being delivered by a TCP connection using only the information provided by TCP/IP headers.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3158", "sentence1": "Currently, communication between nodes is made in plain text because there is no encryption process.", "sentence2": "in this paper, we have performed in a private IP environment, but it can attack any number of URIs only in public IP environment.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3159", "sentence1": "There is no authentication procedure for communication between master and node.", "sentence2": "if the attacker matches only the environment variable ROS_MASTER_URI with the IP address and port number of the master, communication with all nodes connected to the master is possible, which can be exploited for various attacks.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3160", "sentence1": "However, since no separate encryption processing is performed, the communication history is directly exposed to the attacker.", "sentence2": "if an attacker catches a packet in the middle, the robot can be controlled by manipulating the communication details freely from the remote.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3161", "sentence1": " Currently, there is no authentication procedure for ROS network in constructing distributed environment.", "sentence2": "it is necessary to introduce a user request system in the process of communicating with the first master.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3162", "sentence1": "In the ROS, data messages exchanged between the master and the node, called a Bag, Bag file can be used to play this back when you need it and repeat the same action to the Robot.", "sentence2": "the configuration environment for the ROS authentication scheme is ROS Indigo version, and Fig. 1 shows the Raspberry pi 2 environment running the master, and Fig. 2 shows the attacker's environment. ", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3163", "sentence1": "Completion of connection with KOBUKI and Raspberry Run the keyop package in the victim s environment to control KOBUKI with the keyboard.", "sentence2": "we have analyzed the vulnerability of the most representative robot platform .", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3164", "sentence1": "In the RSA-SW scheme, a maximum of 500 POR responses can be performed within the same time lapse, resulting in a modest peak throughout of 10 operations per second.", "sentence2": "the provider can scale approximately 4 times better in SPORT when compared to the RSA-SW scheme of [30].", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3165", "sentence1": "Notice that a static adversary can still collude with corrupted users but these are under the control of the attacker right from the start.", "sentence2": "the attacker can create both honest and corrupted users but cannot corrupt honest users.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3166", "sentence1": "The task of the environment is to play the role of all honest users and to challenge the adversary.", "sentence2": "whenever we say that an honest user executes a certain protocol, we mean that the environment honestly executes the protocol on behalf of this user.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3167", "sentence1": "As defined in Section 2.1, the system model is an abstrac­tion of the system under attack and defines all the possible interactions of the attacker with the system.", "sentence2": "the first step for the definition of a general system modeling technique for any CPS is the identification of the relevant aspects and components of the system.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3168", "sentence1": "The tool can automatically security argument graphs describing: the workflow of a CPS (i.e., how the system provides its functionalities), the security goals, and an attacker model.", "sentence2": "we first present their terminology (relevant for our work) and then we use those definitions to informally describe their attacker model framework.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3169", "sentence1": "In general, the attacker framework proposed in [31] can be applied to several security techniques, e.g. from risk analysis to model checking.", "sentence2": "in this work we focus on modelchecking based techniques.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3170", "sentence1": "There are a number of different work in the literature for the definition of a system model for CPS, e.g., [15].", "sentence2": "the problem has been often studies from an engineering point of view, with no (or little) attention for the security aspects [12].", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3171", "sentence1": "In our notation, a system model is formally repre­sented by the following main concepts.", "sentence2": "in this work we focus on model­checking based techniques.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3172", "sentence1": "In order to concretely exploit on the real system the attacks (or attack traces) found with a formal assessment, one has to concretize those attacks.", "sentence2": "one has to bridge the abstraction gap between the formalized system and the real system.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3173", "sentence1": "For the same reason they did not have access to the L0 ring.", "sentence2": "the analysis tool gives the user the possibility to tune the parameter used during the analysis.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3174", "sentence1": "There­fore, only the communication between the upper-entities can be seen by the attacker.", "sentence2": "by giving/removing to the attacker knowledge encryption and decryption keys, it is possible to grant/restrict the attacker to a sub-part of the network of the CPS.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3175", "sentence1": "Information-theoretically hiding commitments can only be computationally binding [4].", "sentence2": "commitments are renewed on a regular basis.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3176", "sentence1": "They utilize time­stamp chains to prolong the validity of complexity-based dig­ital signatures thereby protecting integrity and authenticity for any length of time.", "sentence2": "these solutions prohibit long-term confidentiality protection.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3177", "sentence1": "Password policies were originally designed to make users pick stronger passwords.", "sentence2": "research has shown that they often fail to achieve this goal.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3178", "sentence1": "Those password composition policies make users come up with new passwords in case their preferred option is disqualified.", "sentence2": "users continue to create new passwords until they have a suf­ficient set to fulfill most policies [8].", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3179", "sentence1": "Thus, neither complex­ity nor length requirements prevent password reuse.", "sentence2": "only the diverse range of character sets can be a limiting factor regarding non-modified reuse of a single password.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3180", "sentence1": "Mix-based countermeasures for Tor can be considered satisfactory if they reduce the success of attacks on Tor, e.g., passive traffic analysis attacks like confirmation, and preserve acceptable performance rates at the same time.", "sentence2": "on average nearly half of Tor circuits are vulnerable to attacks that allow for the de-anonymization of users.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3181", "sentence1": "For state-level adversaries or in case of collusion, this can be increased to a coverage of up to 85 %.", "sentence2": "in contrast to Tor, classical mix networks [4] and anonymous remailers [5] disrupt metadata relations by adding arti.cial delays during the transmission process.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3182", "sentence1": "A file is written to the shared repository if and only if the writer successfully distributes the file’s tokens onto at least t owners’ accounts.", "sentence2": "a write access granted to the shared repository is equivalent to write access to at least t of the owners’ accounts.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3183", "sentence1": "Similar to other MLE instantiations, convergent encryption suffers from a number of attacks such as dictionary attacks [7]; here, if the cloud server can predict the content of a message/file, then it can derive the corresponding encryption key and decrypt the message.", "sentence2": "one generic mitigation of side channels in the cloud is frequent migration, i.e. switching to a separate set of the underlying resources.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3184", "sentence1": "For instance, performing access control using data encryption alone is not sufficient, since shared ownership demonstrates the need for higher-level isolation mechanisms.", "sentence2": "shared ownership should be resistant to key leakage and collusion attempts.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3185", "sentence1": "Physical keys and scannable personal IDs are examples of possession-based authentication.", "sentence2": "the Leven­shtein distance indicates how close the guess is to the origi­nal password [1, 19, 38].", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3186", "sentence1": "This is due to the time it takes to perform a gaze gesture, look to the front again, then per­form another gaze gesture.", "sentence2": "guesses against GazeEnd are the closest to the actual password.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3187", "sentence1": "For the proposed protocol we make use of a secure key derivation function (KDF) to derive cryptographic keys using a secret input data, e.g., a password, with sufficient min-entropy.", "sentence2": "we use a KDF to generate keys for an IND-CPA-secure symmetric encryption and a strongly-unforgeable digital signature scheme.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3188", "sentence1": "Motivated by the aforementioned research findings over respiratory sounds, we hypothesized that acoustic features derived from different breathing patterns, measured by a microphone sensor, should provide an effective approach for authenticating people. ", "sentence2": "we needed to discover the key features (e.g., amount of air, forcefulness, time duration of inhalation and exhalation, etc.)", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3189", "sentence1": "Physiological Context: Our studies in Section 5.4 demon strate the robustness of the system under certain short-term extrinsic contextual factors, such as physical activity.", "sentence2": "breathing-related features are also likely to exhibit longer timescale changes, due to intrinsic physiological changes, such as aging and decrease in the lung capacity.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3190", "sentence1": "These works report a very low EER (Equal Error Rate) in the range of 1%-5%.", "sentence2": "all these works fail to test the security of touch gestures in the presence of sophisticated attacks such as shoulder surfing or video based observation attacks.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3191", "sentence1": "This is because for the VBS video, the various partition modes can be fully utilized to quantize the quantization distortion, and it can also be considered to increase the feature diversity.", "sentence2": "a compact distribution will be helpful to capture more statistical information with a small threshold.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3192", "sentence1": "2.2 Efects of MV Based Steganography on Video Statistics According to current MV based steganography [1, 3, 6, 10, 27, 29], one or two components of a MV can be modified during e emmbeding.", "sentence2": "the features in [20] and [26] are based on the first-order distributions (histograms) of NMVDs and the joint distributions of NMVDs.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3193", "sentence1": "The difference calibrated features can be completely derived from Cartesian calibrated features, but not vice versa.", "sentence2": " the Cartesian calibrated features contain more discriminative information than difference calibrated features.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3194", "sentence1": "Stego signal is a high-frequency noise modulated by content.", "sentence2": "to force the network to pay attention to this signal (and to converge), images need to be pre-filtered to incease the SNR between the signal of interest (the weak stego noise) and the noise – the image content.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3195", "sentence1": "For the first task, 11 children were able to request login without help, while remaining completed with some help.", "sentence2": "apart from design improvements and suggestions for addi­tional features, the user studies provided additional insight.", "label": "neutral", "domain": "Security and privacy"}
{"id": "test_3196", "sentence1": "We found that participants were engaged and able to envision themselves using the product when they could interact with a prototype, enabling them to consider additional features and consider what would make the most sense for their family.", "sentence2": "we found that children were more engaged when interacting with the prototype versus when being interviewed.", "label": "entailment", "domain": "Security and privacy"}
{"id": "test_3197", "sentence1": "The UAF service authenticates a device and then trusts local authorization to verify the user from that device.", "sentence2": "the Universal 2nd Factor (U2F) [30], offers secure second factor authentication.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3198", "sentence1": "Completely eliminating control-flow inference attacks that we model in Section 3 is extremely challenging.", "sentence2": "our study of SSL/TLS implementations’ sensitive control-flow vulnerabilities is not completely addressed by any of these specific defense techniques.", "label": "reasoning", "domain": "Security and privacy"}
{"id": "test_3199", "sentence1": "We need to adopt neural network algorithms within limitations of homomorphic encryption.", "sentence2": "the computation performed over sensitive data by neural network algorithms is very complex and neural networks cannot simply be translated to encrypted versions without modification.", "label": "contrasting", "domain": "Security and privacy"}
{"id": "test_3200", "sentence1": "Based on one of the latter, we analyze the current IDE support for receiving code changes, finding that historical information is neither visible nor easily accessible.", "sentence2": "we devise and qualitatively evaluate BE L L E V U E, the design of an IDE extension to make received changes always visible and code history accessible in the editor.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3201", "sentence1": "Despite this, the TS stack is linearizable with respect to stack semantics.", "sentence2": "theorem 1 points towards fundamental constraints on the structure of concurrent stacks.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3202", "sentence1": "For example, the rule for an if expression type checks each branch under the assumption that the condition is true or false.", "sentence2": "it will then examine the congruence equivalence class of these expressions to see if they contain any suitable values any value v is suitable for a, a function value rec f x.a0 for f, and a value headed by a data constructor for b and then unfold the resulting expression (rec f x.a0) v. (If there are several suitable values, one is selected arbitrarily).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3203", "sentence1": "The total-store-ordering semantics [23] provides that all instructions are executed in program order (or, more precisely, cannot be observed to be executed out of program order), each write is visible either globally or only to its own thread, and writes become globally visible in program order.", "sentence2": "visibility and execution-order edges compile away to nothing, and pushes can be implemented by mfences.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3204", "sentence1": "Alternatively, rather than restricting the setting in such ways, we can instead constrain the programs under consideration.", "sentence2": "we can restrict ourselves to so-called recursive coalgebras [6] those that by definition yield unique solutions to the hylo equation, whatever the algebra.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3205", "sentence1": "More formally, suppose we have some restricted programming language T such that programs of this language can run on the decentralized components.", "sentence2": "there is a subtle local-versus-global issue that is a frequent cause of errors in SDN implementations.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3206", "sentence1": "This problem can be difficult and error prone even for experienced programmers.", "sentence2": "there is a subtle local-versus-global issue that is a frequent cause of errors in SDN implementations.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3207", "sentence1": "For an integer k, we say that a policy F has k-bounded causality if for any event e and history H, if e is relevant with respect to F and H then it is also k-relevant.", "sentence2": "a policy F has k-bounded causality if whenever an event e is relevant, it can cause a change after at most k additional events.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3208", "sentence1": "Similarly, if a load immediately follows a store to the same location, then it is always possible for the load to get the value from that store.", "sentence2": "it is always possible to remove the load.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3209", "sentence1": "Sakunkonchak et al. [33] apply CEGAR optimizations to software model checking and speed up the search for predicates that make the counterexample spurious. ", "sentence2": " they do not use interpolants and instead search the counterexample for conflicting predicates. ", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3210", "sentence1": "Unlike the programming model in this paper, streaming based programming models are not analyzable nor do they break data dependencies to expose data parallelism like we do in this work.", "sentence2": "streaming languages can only run as fast as hardware can expose a stream.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3211", "sentence1": "In this section we describe how to run k-SFTs on data-parallel hardware.", "sentence2": "we demonstrate an end to end compilation of k-SFTs to a data parallel version capable of exploiting multiple cores via threads.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3212", "sentence1": "Over the course of 6 assignments, students complete substantial portions of core system components, including the memory manager, scheduler, file system, shell, and network driver.", "sentence2": "jOS requires students to program x86 hardware including configuring the page tables, context switching the CPU, handling interrupts, and writing an e1000 network card driver.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3213", "sentence1": "The process ABI includes higherlevel abstractions, such as file descriptors, signals, and network sockets, whereas a VM manipulates emulated hardware abstractions, such as a virtual disk, CPU interrupts, and a virtual network card.", "sentence2": "adding VM support to a legacy OS, such as adding KVM to Linux, requires replacing the system call table with a hypercall table, the ability to trap accesses to privileged hardware, and a PC hardware model, generally running as a separate process.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3214", "sentence1": "This example uses a filter to test how well a class is encapsulated.", "sentence2": "the test looks for public fields that are not declared as const by applying a filter to the getFieldNames method (see Figure 8).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3215", "sentence1": "ParTE [5] is a new refactoring tool built on top of Wrangler and RefactorErl.", "sentence2": "wrangler's API and DSL support for scripting is used in ParTE to build refactorings [6], whereas RefactorErl's program analysis support is used to find parallelisable code candidates.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3216", "sentence1": "In formal notations, we use API to denote { CLAPI }, and operations on APIs are naturally set operations (e.g. API1 - API2 is set substraction, which excludes class declarations in API2 from API1 ).", "sentence2": "we use the notation APIs to denote the source API, CLAPI , and APId to denote the target API, CL'API , respectively.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3217", "sentence1": "We found that for all assignments in each program the types we observed formed a subset of the in­ferred type sets.", "sentence2": "the type analysis was able to infer all types observed at run-time, providing an empirical approxima­tion to soundness.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3218", "sentence1": "An important limitation of our technique is that it requires knowledge of the construction of the higher order attribute.", "sentence2": "when a higher order attribute a is used to construct a new value of a higher order attribute, a path needs to be constructed to indicate in which part of the new AST a ends up.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3219", "sentence1": "In order for our technique to work we therefore require the construction of higher order attributes to be of a restricted form in which only constructors, attribute references, and constants are used.", "sentence2": "pattern matching is not allowed as it would highly complicate dependency analysis.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3220", "sentence1": "Actions have priority according to their order in the list.", "sentence2": "if a process is enabled to execute Action B1, it may not execute any other action; if a process is enabled to execute Action B2, it may not execute Actions B3 through B7, and so forth.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3221", "sentence1": "Splitting the heap into two generations enables the GC to use a run-time-efficient collection strategy for objects that are likely to die (i.e., young objects) and a more memory-efficient strategy for objects that are unlikely to die (i.e., old objects).", "sentence2": "the run-time-efficient Scavenge algorithm (minor GC) is used for the young generation, whereas the Mark and Compact algorithm (major GC) is used if all spaces need collecting.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3222", "sentence1": "Because of the nature of this system being an internal business system, it is possible that the number of unhappy users who do not complain is smaller.", "sentence2": "regardless of the actual factor involved, the number of complaints is a good indicator of user satisfaction.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3223", "sentence1": "In particular, one key constraint in quantum computation is the no cloning theorem [33], which states that a superposition state cannot be perfectly copied from one qubit to another qubit, without destroying the first one.", "sentence2": "in order to use the state of a qubit in another part of the architecture, that qubit must be physically moved.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3224", "sentence1": "Note that the functionality of a device is not broken if it is disabled after finishing one task and enabled before handling the next one, as long as the driver preserves and then restores the device context, respectively.", "sentence2": "this incurs unnecessary overhead due to power-state transition.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3225", "sentence1": "We face a trade off in choosing Tthreshold: with a smaller value of Tthreshold, the software approach can infer in a more timely manner if a device has pending tasks and can capture shorter periods in which a device has no pending task.", "sentence2": "a smaller Tthreshold leads to more aggressive PM; however, to avoid false report of no pending task, Tthreshold needs to be greater than the largest possible interval between register accesses when a device has pending tasks.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3226", "sentence1": "ICEM [48] integrates PM code into locks for device drivers.", "sentence2": "the approach only focuses on a specific class of device drivers ( \"shared drivers\" ) which is uncommon in Linux for modern SoCs, according to our observation.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3227", "sentence1": "Instead, it indicates the device has been busy at least once since the last read of the busy/idle register.", "sentence2": "if the register shows idle, the device must have been idle since the last read of the busy/idle register.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3228", "sentence1": "Following the Spark developers' advice, we use the remaining RAM to mount an in-memory file system tmpfs for Spark's intermediate results.", "sentence2": "these experiments make use of all the available RAM.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3229", "sentence1": "Upon colliding with b, ball a may speed up and collide with c before c can collide with d, invalidating task (c, d).", "sentence2": "a task graph is an acyclic graph whose nodes correspond to some subset of W .", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3230", "sentence1": "Many complex Ziria processing blocks will need to both push and pull data during their execution.", "sentence2": "this vectorization candidate is nevertheless incorrect because c1 could stop after reading the first ain inputs from t s output, leaving t with extra ain values which should have been processed by c2.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3231", "sentence1": "The acceptable data widths of a component may depend in complex ways on the rest of the pipeline (Section 3).", "sentence2": "a programmer needs to jointly optimize all the data widths in a pipeline, which is error-prone and tedious.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3232", "sentence1": "Attaching NVMs directly to processors will produce non-volatile main memories (NVMMs), exposing the performance, flexibility, and persistence of these memories to applications.", "sentence2": "taking full advantage of NVMMs potential will require changes in system software [3].", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3233", "sentence1": "These high-performance storage systems would be especially useful in large-scale data center environments where reliability and availability are critical.", "sentence2": "providing reliability and availability to NVMM is challenging, since the latency of data repli­cation can overwhelm the low latency that NVMM should provide.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3234", "sentence1": "Another problem with fixed parallelism is that it targets all requests equally.", "sentence2": "long requests have a greater impact on tail latency than short ones.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3235", "sentence1": "However, when a request arrives, we do not know its service demand (and it is dif­ficult to predict if the request is short or long [26]).", "sentence2": "we determine demand as the request executes.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3236", "sentence1": "For example, assuming the aggregator has 10 ISNs, if we want to process 90% of user requests within 100 ms, then each ISN needs to reply within 100 ms with probability around 0.99.  ", "sentence2": "for a total latency of 100 ms at the 90th-percentile response time, the response time of each ISN must be at most 100 ms at the 99th-percentile.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3237", "sentence1": "Long requests have over 2 times speedup with 3 threads. ", "sentence2": "short requests have limited speedup, a factor of 1.2 with 3 threads.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3238", "sentence1": "OS designers are experts in how their systems work: they represent the best opportunity to enhance the security of the system.", "sentence2": "by selecting this abstraction, the outer kernel still manages all aspects of the virtual memory subsystem; however, the nested kernel interposes on all pMMU updates, thereby allowing the nested kernel to isolate the pMMU and enforce any other access control policy in the system, such as the one used to protect nested kernel code and data.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3239", "sentence1": "After the preliminary analysis has been com­pleted, the next step is to select targets for testing.", "sentence2": "we aim to select specific instructions in the program under test that are more likely to trigger a bug, so we can guide our testing towards those in­structions.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3240", "sentence1": "Xposed runs only on rooted devices, and it is not possible to install AlarmScope on non-rooted devices.", "sentence2": "xposed allows us to apply the functionality directly, without modifying the system platform, and this enables the smooth implementation of our prototype.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3241", "sentence1": "On the other hand, Pandora, sent out at regular 1-second intervals 166 bytes while sending and receiving during long time intervals large packets.", "sentence2": "we can conclude that Pandora consumed more energy than iTunes because it kept the Wi-Fi radio at high power state for most of the test duration.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3242", "sentence1": "Finally, out of the three, last chunk strategy performs the worst in terms of stall and switch statistics.", "sentence2": "harmonic-based strategy has commensurate stall and switching statistics.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3243", "sentence1": "So in most cases we believe post-deployment testing will be required.", "sentence2": "second, per-maybe strategy defeats the flexibility inherent to the maybe approach and would devolve into the fragile decision-making we are trying to avoid.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3244", "sentence1": "Nowadays the rapid development of technology can be observed in every aspect of human life, with no exception to the educational world.", "sentence2": "implementation of secure online examination system is a hot topic.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3245", "sentence1": "Therefore, distinct EXAM (Best) of Heuris­tic III (c) was not effective because EFL eliminated dupli­cated additional test data before calculating the suspicious­ness.", "sentence2": "we also excluded version 9 of schedule2 because there is no fail TC.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3246", "sentence1": "If consistency-control scheme is designed based on WAL scheme, the file system reflects the updates from the logging area to the file system area, whenever read request is issued.", "sentence2": "since MinL2R reflects the updates to the file system area directly with rollback-recovery scheme, the application can read the requested data from file system area of SCM storage, regardless of logged data.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3247", "sentence1": "Facial expressions can be used to enhance conversation, show empathy, and acknowledge the actions of others.", "sentence2": "robot behavior that includes at least some rudimentary, human-like facial expressions can enrich the interaction between humans and robots, and add to a robot's ability to convey intention.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3248", "sentence1": "Further, these approaches do not provide practical evidence that the particular visualization and/or hardcoded abstraction is actually useful.", "sentence2": "debugging and bug isolation strategies have seen significant attention [15, 27]. ", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3249", "sentence1": " Participants found the possibility to interact “directly” with the data compelling and thought that it allows them to more quickly reconcile their mental model of the structure with the rendering on-screen.", "sentence2": "participants often simple changed the visual layout of the graph in order to better understand the program state.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3250", "sentence1": "In summary the experimental data suggests that visualizing the heap graph of a running program does benefit debugging efficiency (H1) and that our mechanisms for automatic abstraction and concretization alongside the possibility to directly interact with the heap graph are indeed useful.", "sentence2": "our tool outperforms the baseline interfaces in the hard task suggesting that the utility of our mechanism increases with task difficulty.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3251", "sentence1": "The flowlet API is also similar to Hadoop MapReduce one for easy use.", "sentence2": "the internal mechanism is considerably different.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3252", "sentence1": "The option we chose was to remove the key scheduling from the signature to keep a unique signature for both versions.", "sentence2": "this requirement is illustrated on the left hand side of Figure 4 where two consecutive additions involving constant variables can be simplified if they are rearranged.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3253", "sentence1": "The thought process to have a reusable and generic system was implemented with a lot of rigor when the initial core classes were designed.", "sentence2": "the study revealed that the practices are ignored due to various business reasons.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3254", "sentence1": "Conducting regression testing after every change or some changes to the code is a solution to immediately detect source code changes with side effects.", "sentence2": "regression testing is an expensive process because it becomes very time consuming to run a large number of test-cases or an entire test-suite, covering all the functionalities of a large and complex software.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3255", "sentence1": "The MVC pattern advocates separation of concerns between three entities: models capturing app data, views describing the visual appearance and the controller which handles transformations between models and views along with navigation control.", "sentence2": "to encourage rapid adoption, it is more pragmatic to augment existing popular frameworks with new abstractions, e.g., by creating framework-specific plugins/libaries, instead of creating a new ecosystem from scratch.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3256", "sentence1": "Not surprisingly, most modern web programming frameworks spawned over the last decade, e.g., Ruby on Rails, Django, Symphony, ASP.NET, Backbone.js, enforce some form of MVC-based design.", "sentence2": "our survey shows that these frameworks lack suitable programming abstractions which enable MVC-based design end-to-end.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3257", "sentence1": "We consider a scenario in which development is underway for the release rn, which started with source code from the release rn-1.", "sentence2": "the commits submitted after the release rn-1 are for the release rn.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3258", "sentence1": "In general components follow the idea of information hiding.", "sentence2": "this is not true for the generic model components discussed in this paper.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3259", "sentence1": "Each point in P can be written as a convex combination of the points in vert(P ).", "sentence2": "p is the convex hull of vert(P ).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3260", "sentence1": "The reachability problem for BMS and MMS is fixed parameter tractable, where the parameter is the number of variables.", "sentence2": "it is polynomial for BMS and MMS with fixed dimension d.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3261", "sentence1": "Recent work addressing the stability analysis of controllers at code level has been mainly focused on the controller alone.", "sentence2": "most of the properties of interest of control software lie in how they interact with their environment.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3262", "sentence1": "De-duplication technique is employed to save only one copy of the identical pages across multiple VM snapshots [21, 22, 42].", "sentence2": "these works pay no attention to the availability of snapshot files, or in other words, they expect the underlying file system or storage device to provide the availability.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3263", "sentence1": "HDFS stripes a file into disperse disks, so it is difficult to quantify the loss of file for real disk failures.", "sentence2": "we randomly select pages and mark them as lost when reading them from HDFS.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3264", "sentence1": "Removing page cache from memory state is one compelling approach to reduce snapshot size effectively, and has been well studied [14, 17, 37, 43, 44].", "sentence2": "these works still suffer from several drawbacks.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3265", "sentence1": "For example, any network packets received at server also get recorded as non-deterministic events in the record log and are streamed to the client (so that it can reconstruct server’s state).", "sentence2": " the user is provided the illusion of controlling the server and viewing its console remotely (even though the console is actually that of the client’s replayed VM).", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3266", "sentence1": "This is because LXC's cgroup freezer [24] mechanism begins consuming more time for un freezing a VM, and hence does not scale.", "sentence2": "the number of frozen VMs is limited by this increased transition time, to approximately 300.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3267", "sentence1": "Thus, we believe that within this operating range (up to 250 booted, 300 frozen VMs) all transition times remain largely constant, regardless of the number of booted and frozen VMs.", "sentence2": "we represent the average transition times (within the operating range) as the transition matrix T3×3 (shown in Table 2, where ti, j=time to transition from state Si to Sj).", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3268", "sentence1": "As mentioned earlier, each world also maintains a version number, which increments monotonously with every modification that happens at the world.", "sentence2": "when the state changes from SQ to SQI, the version number would increment by 1.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3269", "sentence1": "The second consequence was that team members who had not been able to attend Kidsteam sessions in-person were now able to do so in the virtual environment.", "sentence2": "one team member's mother had an unexpected change of work schedule and he was unable to attend KidsteamUB sessions.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3270", "sentence1": "widespread use of social networks has lead to an environment where huge amounts of data are created on high and unpre­dictable rates.", "sentence2": "the volume of data, needed to be pro­cessed, demands a distributed approach.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3271", "sentence1": "Two-stroke engines also exist and are sometimes used in aircraft and marine engines, as well as some trucks and motorcycles, due to a more compact size, a lighter weight, and greater efficiency.", "sentence2": "due to exhaust pollution, twostroke engines are losing out to, and being replaced by, fourstroke engines in many applications, especially in modern cars (which is our focus).", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3272", "sentence1": "Since the player fires slower weapons sooner than faster weapons, effectively “leading” the Saucer more, by the time the faster weapon is fired, the outgoing queue has cleared.", "sentence2": " when the second weapon is fired, there is no queue, so no opportunity for optimization.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3273", "sentence1": "On the other hand, post-mortem methods have a number of disadvantages.", "sentence2": "they are unsound, given that they may not point out every implicit communication link in a distributed system.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3274", "sentence1": "Note that the higher the number of  potential vulnerabilities flagged by static analysis, the higher the overhead incurred at runtime.", "sentence2": "for efficiency, it is crucial that static analysis flag as few false-positives as possible.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3275", "sentence1": "As depicted, the heap overhead depends on the applications.", "sentence2": "the global overhead introduced is not a blocker issue as it is relatively small in comparison of the total heap size.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3276", "sentence1": "In the case example, the browser session parameters have been abstracted to qualitative levels in or­der to enable less experienced customers to select a variant.", "sentence2": "there is also the possibility to set custom session protection by setting individual session parameters under CustomBrowserSessionValidation.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3277", "sentence1": "Compared to the current state in Magento, the conceptualization was used to record implementation constraints from features and countermeasures to the architectural entities.", "sentence2": "all inconsistencies between the selections could be checked; this aspect is currently lacking in Magento.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3278", "sentence1": "On one hand, security requirements, including high-level security goals and system level specifications, need to be realized in the design.", "sentence2": "security requirement definition relies on having the design in place: without information about the intended solutions, the threat analysis cannot fully identify possible targets of an attack [9].", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3279", "sentence1": "Security variability in SPLs is addressed in a few studies that mostly focus on requirements [21] and goal models [13].", "sentence2": "variability must be managed also in the design: security solutions, such as authentication, authorization and input validation, tend to crosscut both structures and views in the architecture [31, 22].", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3280", "sentence1": "Thirdly, the configurator treats requirements R as the set of rules that a specific product instance must satisfy, stated as KumbangSec instances that must be present in the configuration, or as attribute values that these instances have.", "sentence2": "variability must be managed also in the design: security solutions, such as authentication, authorization and input validation, tend to crosscut both structures and views in the architecture [31, 22].", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3281", "sentence1": "KumbangSec artifact assumes that security variability is implemented as a number of purposefully introduced design tactics.", "sentence2": "a security variant consists of a number of countermeasures, which are implemented as design-level tactics in the product line architecture.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3282", "sentence1": "Existing big data processing frameworks such as Hadoop and Spark require jobs to be submitted as a binary jars/libs rather than runtime functional calls.", "sentence2": "the resulting data after applying a Transformer can be used for machine learning using a Learner or it can represent the extracted knowledge discovered from the original data.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3283", "sentence1": "Data-intensive mobile apps often rely on data located in the cloud.", "sentence2": "access to this data is likely over a lower-bandwidth and multi-hop connection, compared to the higher-bandwidth, single-hop connection that exists between a mobile device and a surrogate.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3284", "sentence1": "This is because the Offload Server becomes a bottleneck as all communication between mobile devices and the surrogate would go through this component.", "sentence2": "some systems that implement Fault Tolerance tactics (Section 4.2) place the responsibility of detecting and managing disconnections in the Offload Client and Offload Server which therefore benefits from the single point of communication of the lat­ter alternative.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3285", "sentence1": "However, if used at all, these techniques use the software architecture only in a very coarse-grained manner.", "sentence2": "it is hard to make accurate predictions using task-based approaches.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3286", "sentence1": "With the supply side of the equation shifted (Figure 2), it would be reasonable to assume that we would already be seeing greater use of data to avail the above benefits in many if not most development projects.", "sentence2": "6 In Bangladesh, mPower has supported several projects where the above principles have been applied, through negotiations with donors.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3287", "sentence1": "They also provided leverage for creating and maintaining strong ties with classmates, by providing a capability they could share in exchange for sighted assistance.", "sentence2": "their unequal access to resources in the classroom also created dilemmas for their relationships with peers, fostering resentment among students without access to these amplifying devices.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3288", "sentence1": "Loop i is a parallel loop and therefore we can redistribute the stores among the threads to ensure coalesced access.", "sentence2": "we can use the i loop in the thread code as an additional source of parallel threads, and compute a new 2D geometry in place of the original 1D one, such that accesses will be coalesced for tx: in this example this amounts to (1) making i the tx dimension, updating bx correspondingly to capture N threads; and (2) make the original tx dimension ty in the transformed code.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3289", "sentence1": "Both threads use the same shared lock fil system->mutex to coordinate the shared access to fil->system->unflushed->spaces.", "sentence2": "in the dynamic execution, the thread always does not update it, if the buffer is explicitly disabled by the user (i.e., fil buffering disabled(space)=TRUE).", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3290", "sentence1": "For other events, the recording strategy of PE R F PL AY is quite flexible, ranging from complete recording to selec­tive recording.", "sentence2": "pE R F PL AY chooses selective record­ing whenever appropriate.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3291", "sentence1": "Therefore, a performance debugging tool is needed to as­sist the programmer in addressing the problem of ULCPs in their code.", "sentence2": "we propose PE R F PL AY, a replay framework to help programmers understand ULCPs in two aspects.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3292", "sentence1": "Pointer analysis in general is difficult for compilers, limiting optimization possibilities.", "sentence2": "in terms of CFG construction, function pointer analysis has distinct advantages over conventional pointer analysis.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3293", "sentence1": "Our observation is that, in general, it is difficult to predict the impact of corunning applications until run-time, as the co-runner applications in the workload may not be known ahead of time.", "sentence2": "compiler-based tile size selection approach could not have accurate prediction on the runtime application behavior.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3294", "sentence1": "More specifically, our reactive tiling tries to deter­mine the ideal tile size (from among a fixed set of tile sizes) under an OS/hardware-given cache allocation and the tile size selected by the application may affect the OS/hardware­based partitioning decision in the next epoch.", "sentence2": "the cache partitioning decision in the next epoch is a function (among other things) the tile size selected by the reactive ap­plication.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3295", "sentence1": "Existing cache-oriented compiler optimizations such as tiling are largely agnostic to such cache partitioners.", "sentence2": "they are not aware of the existence of such partitioners and therefore cannot adapt to it dynamically.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3296", "sentence1": "Much higher energy and power savings are seen (up to 22% and 41% respectively) in some cases.", "sentence2": "the savings for some of the benchmarks are limited by the current implementation of the optimization, rather than the technique itself.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3297", "sentence1": "The last writer slice at C for log type shows that it was written last by Thread 1 at A, in the log rotation code, which should be atomic.", "sentence2": "the last writer slice shows the programmer the need for log rotation to be atomic with respect to the insert code to prevent this failure.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3298", "sentence1": "When the failure occurs, Thread 1 is at the assertion in append().", "sentence2": "thread 2 may have completed its erase and moved on to some unrelated code.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3299", "sentence1": "Our approach is based on observing this IR-to-assembly translation, and testing if the two are semantically equivalent.", "sentence2": "for each IR-to-assembly translation used by the compiler, ArC heck tests if the IR and assembly instructions have the same behavior.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3300", "sentence1": "For each IR operator, we also specify constraints relating its input and output, and propagate these constraints on the output of an operator to its inputs.", "sentence2": "we use these constraints to generate inputs that will yield the desired output.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3301", "sentence1": "Our approach, called ArCheck (Architecture Specification Checking), leverages the structure of a modern compiler to achieve architecture independence.", "sentence2": "today’s compilers consist of front-ends, typically one per sourcelanguage, that translate source-code to a common intermediate representation (IR).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3302", "sentence1": "Previous section discussed some basic ideas on what to save and where to save for the snapshot.", "sentence2": "web platforms allow multiple apps to run on a single JavaScript engine. ", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3303", "sentence1": "Due to the halting problem, we cannot statically prove that all bad functions will eventually be lowered.", "sentence2": "the presented lowering algorithm always reduces non-recursive calls of bad functions.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3304", "sentence1": "All instances of Line 0 run in parallel on three cores.", "sentence2": "core 1 happens to take longer initializing its private memory, delaying execution not only in Core 1 but also in Core 2.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3305", "sentence1": "For example, on x86 hosts, the constant-like nodes could be commonly optimized by the in­structions with 32-bit immediate, while on MIPS hosts that do not support 32-bit immediate, constant-like nodes can only be optimized by converting to move-like and add-like node.", "sentence2": "when migrating HE R M E S to different hosts, distinct host-specific features should be considered to revise the instruction templates (or optimization rules) to improve the potential performance gain.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3306", "sentence1": "Ac­cording to the published results [1], the translation through­put of Peephole is at most 200 instructions per second.", "sentence2": "the average translation throughput of HE R M E S is about 37K instructions per second, which significantly out­performs Peephole.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3307", "sentence1": "For a function call, the related return instruction is emulated as a conventional indirect branch, which requires expensive guest-to-host address translation.", "sentence2": "in spite of different guests, a call emulation is characterized by a constant-like node that generates the Guest Return Address (GRA), and the TRA should be recorded once the GRA is generated.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3308", "sentence1": "Of course, the feature can be added to new machines which are not yet delivered.", "sentence2": "there should be a way to add new features to the machine, even when the machine is already delivered to the customer.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3309", "sentence1": "The manufacturer of the machine cannot offer fully customized products as modifying the components for each customer would be too laborious and would result in extensive software versioning.", "sentence2": "the manufactured machine batches are too small to gain the benefits of the efficiency of the mass-production.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3310", "sentence1": "This is quite natural as implementation of some of the functionalities such as boom kinematics requires special expertise.", "sentence2": "it is advisable that a malfunction of a single feature do not disable the whole system, only the malfunctioning features should be disabled.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3311", "sentence1": "However, if there is no trace of the original roles within the pattern definition, it is possible to modify the elements of copied solution in an arbitrary way after the pattern has been applied.", "sentence2": "the properties of the solution, for instance with respect to safety, might not be valid within the system although a pattern guaranteeing these has been used.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3312", "sentence1": "In the used UML modeler, the targets of the role binding are only shown by their name.", "sentence2": "we added stereotyped dependencies to the elements of the pattern solution (in the right hand side of the figure 6) for illustration purposes we did that for the two software components only, the other elements are bound as well in the model.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3313", "sentence1": "The constraints defined within a pattern use the binding information to validate the bound application model elements. ", "sentence2": " it must be possible to filter available patterns with respect to this classification.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3314", "sentence1": "As a consequence of following the MACRO-MICROFLOW pattern, the overall architecture design activity is driven by a business­process-centric approach.", "sentence2": "all design constructs are related to business processes, which need to be modeled and implemented.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3315", "sentence1": "That is, today pattern mining is a rather informal or ad hoc process of finding patterns in software systems.", "sentence2": "often the pattern author identifies one or more patterns in his own experiences, and then broadens the scope by searching for the identified patterns in related systems and/or looking for related other patterns in the systems under consideration.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3316", "sentence1": "Patterns thus represent a conceptualization of problem solving behavior achieved by abstracting from implementations and the explicit decisions software designers have made in software development projects when dealing with many influential forces that constitute complex problems.", "sentence2": "software patterns to be newly discovered are successful mental design constructs, which have not yet been discovered as such, but are rather represented as unexplored and unconscious knowledge of expert software designers that is hidden in the software implementations.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3317", "sentence1": "Pattern mining is a term used in the pattern community to describe the process of identifying or discovering patterns in existing software systems.", "sentence2": "as patterns describe established knowledge rather than original solutions, each software pattern is associated with a number of known uses where the pattern is used in an existing software system (Coplien, 1996).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3318", "sentence1": "They represent successful problem solving behavior, which corresponds to the concept that a pattern describes how to and why to solve a problem rather than concretely dealing with what to solve.", "sentence2": "they deal with people s behavior to solve complex design problems, and they describe how and why conflicting forces arise and how and why these forces are resolved by a (human) software designer.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3319", "sentence1": "The service model and, eventually, the business capability model may be refined and updated.", "sentence2": "sophisticated (enterprise architecture) tool-support is required for administrating and updating the models and their dependencies.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3320", "sentence1": "For example, the well-known GoF factory patterns allow to specify a concrete (application) class to be instantiated by the injection of a factory object at runtime.", "sentence2": "the algorithm that determines when an instantiation will take place, can be fixed by coding a (framework) routine long before the concrete class is decided at runtime (see Fig. 1).", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3321", "sentence1": "It is often hard to decide whether something is a tactic or a pattern, because there is no precise definition for tactics.", "sentence2": "in this work we simply take a given set of safety tactics from [Preschern et al., 2013] (where a more detailed explanation about safety tactics can be found) and consider every architecture which uses several of these tactics as architectural safety patterns.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3322", "sentence1": "In our framework, we have to deal with components realized in the C language and an application constructed following the component-based paradigm [Crnkovic 2002].", "sentence2": "just experimental but no mature C++ IoC containers are currently present.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3323", "sentence1": "However, debugging deployed WSNs is challenging for several reasons - the remote location of deployed sensor nodes, the non-determinism of execution that can make it dicult to replicate a buggy run, and the limited hardware resources available on a node.", "sentence2": "existing solutions to record and replay debugging in WSNs fail to capture the complete code execution, thus negating the possibility of a faithful replay and causing a large class of bugs to go unnoticed.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3324", "sentence1": "The composition of the request-processor chain reflects the specific responsibilities of a replica and therefore differs between primary and backups.", "sentence2": "the primary participates as proposer in the protocol responsible for the reliable distribution of state transactions [35], while backups act as learners.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3325", "sentence1": "Although close-to-open semantics do not require Hare to ensure data consistency in the face of concurrent file operations, Hare must ensure its own data structures are not corrupted when multiple cores manipulate the same file.", "sentence2": "if one core is writing to a file and another core truncates that file, reusing the file’s buffer cache blocks can lead to data corruption in an unrelated file, because the client library on the first core is still writing to these buffer cache blocks.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3326", "sentence1": "Hare's insight is that exec provides a narrow point at which it is easy to migrate a process to another core.", "sentence2": "the entire state of the process at the time it invokes exec is summarized by the arguments to exec and the calling process's open file descriptors.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3327", "sentence1": "When the file descriptor is not shared between processes (\"local\" state), the client library maintains the file descriptor offset, and can perform read and write operations without contacting the file server.", "sentence2": "if multiple processes share a file descriptor (\"shared\" state), the offset is migrated to the file server, and all read() and write() operations go through the server, to ensure consistency.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3328", "sentence1": "A task that exceeds its memory limit will be the first to be preempted if resources are needed, regardless of its priority, so it is rare for tasks to exceed their memory limit.", "sentence2": "cPU can readily be throttled, so short-term spikes can push usage above reservation fairly harmlessly.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3329", "sentence1": "The Omega architecture was designed to support multiple distinct workloads that have their own application-specific RPC interface, state machines, and scheduling policies (e.g., long-running servers, batch jobs from various frameworks, infrastructure services like cluster storage systems, virtual machines from the Google Cloud Platform).", "sentence2": "borg offers a \"one size fits all\" RPC interface, state machine semantics, and scheduler policy, which have grown in size and complexity over time as a result of needing to support many disparate workloads, and scalability has not yet been a problem (§3.4).", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3330", "sentence1": "In this paper, we propose a fine-grained monitoring mid­dleware providing real-time and accurate power estimation of software processes running at any level of virtualization in a system.", "sentence2": "our solution automatically learns an application-agnostic power model, which can be used to estimate the power consumption of applications.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3331", "sentence1": "Tango targets a heterogeneous environment in which different replicas have different strengths and weaknesses.", "sentence2": "it is often beneficial to switch the role of leader as the application enters different phases of execution, and Tango is designed to identify such opportunities and switch appropriately.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3332", "sentence1": "The differences in Figures 1 and 2 illustrate that when the server is the leader, it can externalize the UI state on behalf of the client.", "sentence2": "when the client is the leader, only the client will generate UI output.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3333", "sentence1": "Time Warp [24] improved distributed simulation performance by speculatively executing computation on distributed nodes.", "sentence2": "table 1 shows the server's processing time.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3334", "sentence1": "Sony, Nvidia, Amazon and OnLive are among the providers that currently offer cloud gaming services [1 3, 10].", "sentence2": "cloud gaming faces a key technical dilemma: how can players attain real-time interactivity in the face of wide-area latency?", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3335", "sentence1": "Somewhat surprisingly, having test players different from training players only marginally impacts prediction performance as long as test and train players are of similar skill level.", "sentence2": "we chose to use a single model per coarse-grained skill level (novice or experienced) which is agnostic of the player.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3336", "sentence1": "On the other hand, this approach identified 684 new ad module occurrences (7.07% of all 9,676 occurrences), of which only 55 (0.8%) have requested new permissions, compared to ad libraries identified using the non-call-graph method.", "sentence2": "in general, this approach increases cost significantly while providing diminishing returns.\n", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3337", "sentence1": "A model derived by WI can be used to generate artificial workloads that are close to the actual workload but may have random mutations/omissions.", "sentence2": "we can now define the high-level problem solved by WI: Given an input sequence T that we will treat as training data, we wish to compute k and a Markov model Mθ(k) ∈ Mk such that Mθ(k) is a “good” abstraction for the process that generated T.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3338", "sentence1": "The vast majority of slow SSD reads happen while the SSD is in the process of servicing segment write requests.", "sentence2": "we try to avoid writing to more than two SSDs per ECC group at the same time, and treat SSDs that are in the process of writing data as though they have failed.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3339", "sentence1": "The key idea of both techniques is to skip some input data (time-adaptive algorithms) or expensive computations (approximate computing) so as to produce approximate results within the required processing time.", "sentence2": "our framework performs full computation over the synopsis (statistical aggregation) of entire input data.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3340", "sentence1": "In all cases, schedules with proposed rules perform better than the baseline.", "sentence2": "1024 byte value size, rule 1 performs 41%, rule 2 performs 1% and rule 1+2 performs 44% better.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3341", "sentence1": "We are not satisfied with this rough classification and apply a text mining approach to quantify the promoted benefits on a more detailed level.", "sentence2": "we start to count each occurring pair and triple of words in the collected benefit items. ", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3342", "sentence1": "When it comes to the working conditions of software engineers, employers are confronted with the universal mix of extrovert/introvert and sensitive/intuitive people who decide extremely analytical where to be hired.", "sentence2": "recruiters run into a challenge of compensations and benefits as well as in the public reputation of a valuable corporate culture.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3343", "sentence1": "We also compare our results with those of Oracle Solaris Studio 12.3 compiler [15], the only compiler to the best of our knowl­edge that implements some OpenMP correctness checking as well.", "sentence2": "since Sun Studio only supports OpenMP 3.1 it cannot handle issues related to tasks dependences.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3344", "sentence1": "The variable construct defines 3 Genesis variables sampled from my_vars.", "sentence2": "every time the feature computation is processed, the variables dest, src1 and src2 are sampled to select 3 variables from my_vars1 to my_vars5, which are substituted into the code snippet.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3345", "sentence1": "automate the generation of vector­ized and multithreaded linear transform libraries, providing users with optimized code for this domain of applications.", "sentence2": "genesis offers the flexibility to express programs in any domain, but does not optimize them, although optimiza­tions can be manually expressed with Genesis constructs.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3346", "sentence1": "Database systems use the journaling techniques in order to guarantee database consistency even at system crashes.", "sentence2": "such duplicated write operations shorten the lifetimes of NAND flash storage devices.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3347", "sentence1": "Most of the research found checks for consistency only a given state of SPL.", "sentence2": "we believe it should be interesting to have more initiatives on accelerating the discovery of inconsistencies.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3348", "sentence1": "Objective: In this paper we present our findings from a case study to give a deeper understanding of the challenges and problems of user involvement during software development.", "sentence2": "therefore it is not necessary that the users who are involved in the project should also participate and perform activities.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3349", "sentence1": "From the SME's perspective, there was a feeling that expectation of the degree and level of her involvement was not clearly articulated at the outset.", "sentence2": "the responsibilities of the ECC manager (who was conducting all the requirements elicitation and design workshops) were not clarified.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3350", "sentence1": "The very small difference in the control groups is surprising to us because we have assumed that the quantity of correct retrieved elements is significantly lower in the larger system compared to the quantity of correctly retrieved elements in the rather small system.", "sentence2": "the participants of the control groups from the both experiments performed rather poorly.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3351", "sentence1": "The results pointed out in Section 5 show that the null hypothesis Ho1 can be rejected.", "sentence2": "the first experiment took place with 51 students of the software architecture course; the second experiment was conducted with another 56 students of the same course.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3352", "sentence1": "Furthermore, since the re­spective variant should be grounded in the reference model instead of the V-Modell XT Bund, required variability oper­ations had to be defined in a new metamodel variant.", "sentence2": "the development approach of the V-Modell XT Bund was replicated by another team in another context.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3353", "sentence1": "An exception is the operation type ReplaceTaskDescription : The original operation (defined in the metamodel MM 1.4B ) remains unused, but the copied operation (defined in the metamodel MM 1.4Bw ) was used.", "sentence2": "the original operation remains in the set of unused operations, even though an exemplar is present in the analyzed process variants.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3354", "sentence1": "The GRADE system initially categorizes evidence concerning study design by assigning randomized experiments a high grade and observational studies a low grade.", "sentence2": "by considering the quality, consistency, and directness of the studies in the evidence base, the initial overall grade could be increased or decreased.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3355", "sentence1": "When it comes to the papers which refer checklists from the guidelines, we found many papers developed their quality assessment checklists based on previously developed and used quality assessment checklists that can also serve as some sort of guidelines.", "sentence2": "compared to SLRs, it was not expected that many SMSes would have performed the quality assessment of the papers included in the SMSes.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3356", "sentence1": "However, after extracting the data from these SLRs, we found that 17 SLRs that did not report their quality assessment criteria inside the paper or somewhere available to public.", "sentence2": "we divided the 127 SLRs into two lists: 110 SLRs with explicit quality assessment criteria reported, listed in Appendix A, and the other 17 papers (without sufficient information about the quality assessment claimed to have been used) are listed in Appendix B.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3357", "sentence1": "Studies indicated that the DPM method was one of the best Carve-fitting methods.", "sentence2": " the principles of DPM method are quite different to the original CRC method, and as a matter of fact, this method is not a capture-recapture method.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3358", "sentence1": "In recent times, however, there has been growing unease regarding the quality of the Android platform [3].", "sentence2": "kumar Maji et al. [13] studied issues reported for four early versions of the Android OS (versions 1.1, 1.5, 1.6 and 2.0) and found most defects to be present in the application layer. ", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3359", "sentence1": "As usual, each research question motivates some data extraction.", "sentence2": "the instrument was a spreadsheet, where each column represents a piece of information that had to be extracted from the studies.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3360", "sentence1": "It is important to mention that 76 of these 458 studies use support mechanisms only to data analysis and research validity, without applying any mechanisms to guide the planning and the conducting of empirical strategy.", "sentence2": "if we consider the studies that use at least one as reference to aid planning and conducting the empirical strategy, we have only 382 studies, in other words, only 43% of the full papers published in EASE, ESEM, and ESEJ apply guides to their empirical methods.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3361", "sentence1": "The rationale was that the testing effort might be a proxy measure of the claimed effects of TDD (improved quality and productivity), due to its central role in this technique [6].", "sentence2": "the authors claim that, since TDD developers are believed to put more effort in unit testing when com­pared to test-last developers, the effects on external quality and productivity are likely to become evindent when an high level of testing effort is observed.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3362", "sentence1": "No single source document of guidelines for mandatory log events exists to provide a comprehensive overview of what a software engineer should log.", "sentence2": "for our current work, we present a methodology for identifying mandatory log events based on natural-language artifacts specific to the individual software system.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3363", "sentence1": "Logging everything often introduces resource and performance [4] issues.", "sentence2": "heuristic 2 covers the most verb-object pairs (35%); Heuristic 2 and Heuristic 3, together, cover approximately 47% of verb-object pairs; Heuristics 2-4, together, cover approximately 57% of verb-object pairs, etc.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3364", "sentence1": "For example, use create instead of make , use edit or add (as appropriate) instead of indicate , and create a blog entry instead of blog .", "sentence2": "in use-case based requirements, sentences are allowed to freely follow any grammatical structure and pick from a larger variety of verbs, since use-cases are not constrained to describing only what the system shall do.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3365", "sentence1": "It is assumed that the upper level controllers are connected to the network with wired connection.", "sentence2": "it is considered that there is no point of attack between the network and the controller.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3366", "sentence1": " As technology evolves and new programming paradigms are born, Design Patterns evolve too: patterns for distributed and parallel computing or Cloud oriented and High Performance architectures have been defined to meet current applications requirements.", "sentence2": "here we focus on Cloud Patterns, that can be seen as a recent evolution of Design Patterns but especially defined for Cloud platforms and environments.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3367", "sentence1": "When a Design Pattern is detected, it is possible to identify a well defined set of software requirements it satisfies: being able to determine the Cloud Pattern that fulfills the same (or similar) requirement can lead to an automatic porting of the examined software system to a Cloud platform.", "sentence2": "using patterns characteristic expressed through a semantic based formalism, we propose a tentative procedure to assess the equivalence between Design and Cloud patterns, aiming at providing a set of possible mappings between the components belonging to different patterns.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3368", "sentence1": "The main problem is that it could be rather impossible to automatically analyse the content of the patterns described in this way and to compare them with others.", "sentence2": "adding negative consequences decreases the evaluated score, since they can have a negative effect on the migrated application, if the Cloud pattern was chosen for the mapping.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3369", "sentence1": "At the light of these insights, it makes sense for a developer to switch from continuations to Promises.", "sentence2": "the refactoring of existing code bases might be an operation impossible to carry manually within reasonable time.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3370", "sentence1": "Such an \"all-or-nothing\" [15] data placement policy, i.e., placing the whole input of one job in one tier, is likely to yield good performance.", "sentence2": "we further vary the partitioning by increasing the fraction of input data on faster ephSSD (Figure 5(b)).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3371", "sentence1": "While the central location that acts as a hub is often located in a well-provisioned data center, the resources are typically limited at the edge locations.", "sentence2": "the available WAN bandwidth between the edge and the center might be limited.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3372", "sentence1": "This 1) greatly reduces the number of operations related to managing the OMS, 2) reduces the amount of information that needs to be cached in the processor TLBs, and 3) more importantly, enables the memory controller to completely manage the OMS with minimal interaction with the OS.", "sentence2": "naïvely implementing the second step will involve a TLB shootdown for the corresponding virtual page.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3373", "sentence1": "Finally, assuming each cache tag entry requires an additional 16 bits for each tag, across a 64KB L1 cache, 512KB L2 cache and a 2MB L3 cache, the cost of extending the cache tags to accommodate a wider physical address is 82KB", "sentence2": "the overall hardware storage cost is 94.5KB", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3374", "sentence1": "In contrast, with our framework, the system can potentially use an overlay for each virtual page to store metadata for the virtual page instead of an alternate version of the data.", "sentence2": "the Overlay Address Space serves as shadow memory for the virtual address space. ", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3375", "sentence1": "Many libraries in the HPC field encapsulate sophisticated algorithms with clear theoretical scalability expectations.", "sentence2": "according to this method, we first calculate clock differences relative to the first process, and then set a time window relative to this process in which every process should start the operation.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3376", "sentence1": "For the test cases that have known analytical models, like MATMUL (O(n3) FLOPS, O(n3) Loads, O(n2) Stores, etc.), it is easy to validate the generated performance models against the source code, as shown in the previous section.", "sentence2": "(See Listing 2 for examples where COMPASS detected strided accesses and SIMD FLOPS.)", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3377", "sentence1": "Furthermore, white boxes denote the stencil group exe­cution times computed as the sum of the maximum between stencil execution times and communication times.", "sentence2": "we estimate the execution time ts of a stencil's that performs cs floating point operations as the time needed to compute the stencil without considering any communication cost.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3378", "sentence1": "Our performance model shows that our kernels, like most stencil computations, are heavily memory bandwidth limited.", "sentence2": "the correlation factors of 1.5x respectively 1.6x can be attributed to the fact that the kernels attain only a fraction of the peak main memory bandwidth.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3379", "sentence1": "We put all stencils that correspond to a specific tiling hierarchy level into brackets.", "sentence2": "a hierarchical tiling results in a nested bracket expression with the outermost bracket term representing the bottom of the tiling hierarchy.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3380", "sentence1": "Each task in the TDG has a list to include its predecessors (plist ).", "sentence2": "heat diffusion shows higher overall improvement over the different configurations.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3381", "sentence1": "In this regards, they are not the same as the CPUs and therefore when we divide work across the nodes, and between the CPU and the accelerator, we need to partition such that the accelerators only require data trans­fers with their hosts.", "sentence2": "the task graph for the problem is such that the tasks assigned to any accel­erator have data-dependencies only with their host CPUs.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3382", "sentence1": "Consid­ering the Xeon PhiTM s higher core count the situation be­comes worse, as there are approximately 134 MB of RAM for every Xeon PhiTM core, compared to 2 GB of RAM for each CPU core.", "sentence2": "the Xeon PhiTM is most effectively used as a large SMP node, programmed using a shared-memory paradigm.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3383", "sentence1": "Currently, ASPaS supports two variants of bitonic merging networks, because their corre­sponding permutation operators can symmetrically and com­pletely utilize all elements in each input vector.", "sentence2": "other merging networks, e.g., odd-even network, cannot re­arrange all elements and need more masks to hide irrelevant elements in each concurrent step, leading to additional over­head.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3384", "sentence1": "Currently, ASPaS supports two variants of bitonic merging networks, because their corresponding permutation operators can symmetrically and completely utilize all elements in each input vector.", "sentence2": "after threaded through the wires of the network, these 4 elements are sorted.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3385", "sentence1": "As a result, for every 16*16 elements, the “no-vec” version needs 16*60 = 960 comparators to partially sort the data.", "sentence2": "the code generated by ASPaS only needs 60 vector comparators, and then uses 4 data-reordering operations (Eq.1) for each pair of vectors in the aspas_transpose to transpose the data.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3386", "sentence1": "All previous decisions, which resulted in any kind of design or implementation, constitute a limiting factor constraining the ongoing development.", "sentence2": "current research is aware of the influencing nature of design constraints but does not sufficiently address them, especially not in the context of design decision reasoning during software evolution.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3387", "sentence1": "Solution Repository: The tool has a repository containing solutions, grouped by different categories like architectural styles, design pattern, frameworks, COTS, and refactoring solutions.", "sentence2": "to complete the process, the selected solution ADOutcome from the previous step is implemented in the last step.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3388", "sentence1": "As stated in our previous work [6], the proposed testbed considers game software as inter-connected components, which function as cooperative modules via inter-component message transfer.", "sentence2": "a game application is decomposed into a number of pieces, which either executed in the cloud or the players terminal, according to the status of devices and network quality.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3389", "sentence1": "In other words, a game application is decomposed into a number of pieces, which either executed in the cloud or the players terminal, according to the status of devices and network quality.", "sentence2": "the testbed is required to be capable in perceiving the application runtime environment, making decisions of workload allocation, and facilitating the dynamic inter-connectivity between cloud/terminal components.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3390", "sentence1": "Matrix M is computed using matrix Ci-1 and variance ssi-1 that are carried over from the previous iteration.", "sentence2": "based on our analysis, the most promising PCA approach for large datasets is the probabilistic PCA.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3391", "sentence1": "This hap­pens because MLlib-PCA performs dense matrix operations on the covariance matrix.", "sentence2": "this simple optimization can ben­efit several other machine learning algorithms.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3392", "sentence1": "In order to extend SecLoc to support data re-storage, we extend SecLoc as follows: 1) After the initial storage at state t0, for each time of data re-storage, we allow users to specify a new (random) state and encrypt data at this new specified state.", "sentence2": "as long as the storage stage occurs, a new state would be involved.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3393", "sentence1": "Similar to the black detection problem in the Lego assembly task, detecting white paper is not as easy as looking directly at the RGB values.", "sentence2": "we use an approach similar to the board localization approach in the Lego Assistant to reliably detect the easel surface (Figure 6(b)).", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3394", "sentence1": "However, as seen in Figure 6(e), this detection can be noisy because of shadows and creases.", "sentence2": "that would be necessary, for example, if we were to tackle the far more challenging and difficult task of enabling a blind person to play.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3395", "sentence1": "Likewise, the target address calculation for multiple ToCs to the same target address can be done once.", "sentence2": "many redundant target address calculations can be eliminated.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3396", "sentence1": "A prepare-to-branch (PTB) instruction has been previously pro­posed [3].", "sentence2": "note that the SP architecture eliminates the need for a RAS since the return address is known at the point of the return ToC.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3397", "sentence1": "The approach used in KESO's alias and escape analysis is based on the work of Choi [10].", "sentence2": "behavior, results, and features of the analysis for stack allocation are similar.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3398", "sentence1": "For each run of the optimization pass, allocations are propagated at most a single level up against the di­rection of the call hierarchy.", "sentence2": "running the pass multiple times will increase the maximum scope extension level.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3399", "sentence1": "Again, the LLVM compiler's auto-vectoriser fails to exploit any vectorisation opportunities.", "sentence2": "the port produced by FREE RIDER substantially outperforms the baseline implementation.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3400", "sentence1": "However, what we have observed is that industry has devel­oped its own set of VM tools, potentially reducing technol­ogy transfer from academia, which is vital for any research field.", "sentence2": "in order to contribute to the establishment of VM tools standardization at architectural level, we strongly believe that reference architecture concepts can support such an establishment.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3401", "sentence1": "They contain essential knowledge for developing products encompassing: the knowledge for designing concrete architectures of systems of a given application domain; development best practices; and software elements.", "sentence2": "this paper discussed the importance of an architectural standardization for variability management tools based on reference architectures in order to provide several benefits at developing these tools, such as, promoting interoperability and integration, increasing reuse of projects experiences, and establishing development best practices.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3402", "sentence1": "one should determine the appropriate formalism level for describing domain elements (e.g., legis­lation, standards, quality requirements, and system com­pliance), infrastructure elements (e.g., software elements, hardware elements, and architectural styles), application el­ements (e.g., constraints and functional requirements), and crosscutting elements (e.g., domain terminology, external communication, decisions, and internal communication) of reference architectures for embedded systems.", "sentence2": "a formal description can support the automated comparison and/or selection of components for instantiating these refer­ence architectures.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3403", "sentence1": "Considering that components can be used for realizing reference architectures, we also investigate how they could be expressed in such a formal description.", "sentence2": "we focus on embedded systems due to the sophistication required for systems in this domain.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3404", "sentence1": "Model-Driven Development (MDD) is a software development approach where models are first-class citizen [24].", "sentence2": "a special type of abstract model, so-called metamodel, is essential to this approach.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3405", "sentence1": " In order to discuss how component-based development can contribute to the definition and use of reference architecture in practice, in this position paper, we outline some fundamental characteristics of components and composition and posit their relevance to reference architecture.", "sentence2": "compared to objects and architectural units, mapping features to/from encapsulated components is relatively straight forward, since such components have no external dependencies, and are therefore very loosely coupled (via coordination).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3406", "sentence1": "A component model defines components and their com­position.", "sentence2": "rather, software units are usu­ally glued together, albeit often using very sophisticated programming frameworks.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3407", "sentence1": "When reaching this level of capacity the number of cores per node is expected to increase significantly [6] while node memory capacity and I/O latency and bandwidth are expected to not be able grow at the same pace (due to power consumption and technology limitations respectively).", "sentence2": "the gap between CPU and I/O capacity is expected to increase, and I/O to become a performance cap for certain applications.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3408", "sentence1": "The software defined execution environment of a composite solution can be very complex.", "sentence2": "solution operators levy requirement on the execution environment, for example, scaling behavior, access rules, compliance policies.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3409", "sentence1": "When a client is running on Azure, we see much more performance improvement because Azure disk performance is worse than AWS as we shown in Figure 2.", "sentence2": "we omit the result because there is too much performance vari­ance in Azure disk to be generalized.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3410", "sentence1": "That is, retrieving data located from a nearby DC is not always better than a local DC as Figure 1(d) shows.", "sentence2": "determining the boundary between nearby and remote is an open question.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3411", "sentence1": "We now describe the process for calculating the histogram.", "sentence2": "the method of Cho et al. did not extract the embedded watermark.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3412", "sentence1": "3D printing content is a new form of content being distributed in digital as well as analog domains.", "sentence2": "its security is the biggest technical challenge of the content distribution service.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3413", "sentence1": "The original sphere has radially uniform surface normal vectors in all directions based on the center (See Fig.3(a)).", "sentence2": "in the layered sphere most of the surface normal vectors are changed along the z-axis primary test not the x-y plane (See Fig.3(b)).", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3414", "sentence1": "Also, at any given time, only a small number of applications (up to w) are generally expected to be processed given the DS-Pc.", "sentence2": "whenever the service-queue is processed in our VARSHA framework, the number of times that our region-selection heuristic would need to be invoked is bounded by a relatively small constant integer.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3415", "sentence1": "In [5], authors claimed the first use of probabilistic model checking to perform RAM analysis of satellite systems.", "sentence2": "use of exponential distribution to model both failure and maintenance is a major limitation of their work.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3416", "sentence1": "Open initiatives like openSCAD [16] and openJSCAD [21] are gaining footholds within the scripted-CAD community. ", "sentence2": "these tools allow programmers to produce end-user interfaces for customizing models which allow users to alter features such as width, curves, or text.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3417", "sentence1": "In our static analysis approach, we focus on comparison of the difference between permission and intent usage for each mHealth app.", "sentence2": "to the best of our knowledge, none of the related work proposes the framework of auditing on mHealth apps, which monitors the resource usage patterns of mHealth apps in real-time and triggers alerts to users if abnormal resource usage patterns are detected.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3418", "sentence1": "For example, it is possible to store multiple worksheets, and to include not only data, but formulas, macros and images.", "sentence2": "it is possible to extract not only data but metadata and annotations as well.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3419", "sentence1": "This is useful for statistical purposes, but not for identifying the riskier zones and seasons.", "sentence2": "these visual overviews were co-designed together with a series of stakeholders, taking into consideration their feedback and recommendations.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3420", "sentence1": "Similarly, a distance 1 occurs in the case that there is only a single undecoded neighbor, which can then be recov­ered uniquely.", "sentence2": "a distance of 0 or 1 provides information about the recovery of neighbors that are part of a received linear combination, and thus they can be viewed as an ACK on a group of symbols (i.e., neighbors of the linear equation).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3421", "sentence1": "Not only does the tester not have control over the internals of the system she now also does not have control over the environment in which the system is executed.", "sentence2": "the same test, executed at different times, may cause different loads in the system.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3422", "sentence1": "Ideally, in object­oriented systems, for each production class, we have a re­lated counterpart in the test section.", "sentence2": "the size of the test suite increases linearly with the size of the system.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3423", "sentence1": "We start with 340 apps and unfortunately 42 apps fail to go through due to the same reason as we mentioned above.", "sentence2": "here we only show our experimental results for the rest 298 apps.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3424", "sentence1": "Of special interest to our analysis are the dangerous and signature* permissions, because of their impacts.", "sentence2": "increased system complexity means, in general, not only more functionality and larger source-code, but also more intricate system architectures.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3425", "sentence1": "While it is inherently hard to perform large-scale dy­namic analysis, we acknowledge the limited size of our 100-app corpus.", "sentence2": "obviously, if the verification is not implemented correctly, an attacker could impersonate the server.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3426", "sentence1": "Observe that Algorithm MVC specifies up to K + 1 iterations, each iteration consisting of 3 steps enumerated in the algorithm.", "sentence2": "the algorithm may potentially terminate before completing all K + 1 iterations (due to the check in step 3).", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3427", "sentence1": "First, observe that the four ports of the nodes introduce more choices for the scheduler in every step.", "sentence2": " these new choices, if treated uniformly, result in the same multiplicative factor for both the “positive” (an (l, q0) interaction) and the “negative” (an (l, q1) interaction) events, so the probabilities of the process are not affected at all by this.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3428", "sentence1": "Informal description: Every node u has a unique id idu and tries to simulate the behavior of the unique leader of the protocol of Theorem 1.", "sentence2": "whenever it meets another node for the first time it wants to mark it once and the second time it meets that node it wants to mark it twice, recording the number of first-meetings and second-meetings in two local counters.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3429", "sentence1": "This was very convenient for studying the capability of such systems to self-organize into abstract networks and it helped show that arbitrarily complex networks are in principle constructible.", "sentence2": "this is not expected to be the actual mechanism of at least the first potential implementations.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3430", "sentence1": "Figure 1 suggests a link between the phase lengths and the number of nonperfect time steps.", "sentence2": "every phase of an epoch except for the last one contains only non-perfect time steps.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3431", "sentence1": "One is a Linux kernel; the other is an LWK.", "sentence2": "none are sufficient to include the latest crop of LWKs.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3432", "sentence1": "One approach to making the approach more robust would be to use an exhaustive set of test cases, covering all pos­sible error conditions and program inputs.", "sentence2": "in many cases, particularly for those who use commercial software, the source code for programs is often not available.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3433", "sentence1": "Phase angle differences across different buses in a system are a measure of static stress across the grid and its propensity to instability.", "sentence2": "phase angle differences are required to be monitored with respect to predetermined stability thresholds.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3434", "sentence1": "A global scheduling mechanism was employed for load balancing, decreasing the contention and improving performance.", "sentence2": "the other is the SW2 node (of Guillimin), consisting of two Dual Intel(R) Sandy Bridge EP E5-2670 2.6 GHz CPUs, 8 cores per processor, 8 GB of memory per core, and a Non-blocking QDR InfiniBand network with 40 Gbps between nodes.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3435", "sentence1": "This paper explores the use of transactional memory as an alternative to conventional synchronization mechanisms for managing the pending event set in a Time Warp synchronized parallel simulator.", "sentence2": "we explore the application of Intel's hardware-based transactional memory (TSX) to manage shared access to the pending event set by the simulation threads.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3436", "sentence1": " When the simulation does appear to run normally, it executes slightly faster than the strictly static thread assignment scheme. ", "sentence2": " the instability of this migration scheme made it infeasible to obtain data.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3437", "sentence1": "This way we achieved differentiated configurations in terms of the relation between the event execution granularity, and the granularity of the extra-ticks’ interval (recall this has been set to 100 microseconds).", "sentence2": "the adopted settings allowed us to determine different actual likelihoods for an extra-tick to interrupt an on going event (in fact such a likelihood is higher when the event granularity is greater), which gave us the possibility to assess our time-sharing architecture when changing the likelihood that a higher priority task can be detected as standing while the execution of an event is in progress.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3438", "sentence1": "Then, the model computes the time the car will need to traverse the node, adding traffic slowdowns which are again computed according to a Gaussian distribution.", "sentence2": "the probability of finding a traffic jam is a function of the number of cars which are currently passing through the node.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3439", "sentence1": "Such a reduction, which can lead to improving the memory local­ity and the cache/RAM hierarchy efficiency, can be consid­ered as a reflection of the whole optimization process leading to well-suited tradeoffs between log and restore overheads.", "sentence2": "these literature solutions do not directly tackle the issue of memory-access efficiency in NUMA platforms.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3440", "sentence1": "The sequential execution is a baseline one, which allows us to verify the level of speedup achievable via parallel runs.", "sentence2": "the glibc-based parallel run is a reference one allowing us to assess the effectiveness of the ad-hoc NUMA-oriented memory management support for Time Warp (when compared with standard memory allocation and management facilities).", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3441", "sentence1": "Since no transformation is done, the warehouse retains all the information embedded in the data.", "sentence2": "with that format of data, there is more chances that interesting information can be found.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3442", "sentence1": "Then, few of the most important challenges related to large data warehouse are described.", "sentence2": "the difficulties related to data storage from the sensors are reviewed and many of the data mining problems in the context of Big Data are discussed.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3443", "sentence1": "The main advantages of the method are that the dataset is easy to understand and process and the risk of losing information is limited for simple sensors.", "sentence2": "the story is different for more complex sensing technologies such as ultrasound and laser range scanner.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3444", "sentence1": "Loop transformations are known to be important for performance of compute-intensive programs, and are often used to expose par­allelism.", "sentence2": "even if a compiler is capable of automatically transforming a loop nest to expose parallelism, the resulting code may be complicated and different from the original, making it difficult to understand and/or to maintain for the pro­grammers.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3445", "sentence1": "Loop transformations are known to be important for performance of compute-intensive programs, and are often used to expose parallelism.", "sentence2": "many transformations involving loops often obfuscate the code, and are cumbersome to apply by hand.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3446", "sentence1": "For example, in Couch CMS the backslash and double-quote character is replaced to prevent an outbreak of double-quotes.", "sentence2": "an attacker can terminate the current script tag and start a new JavaScript context that requires no quotes by injecting </script><script>.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3447", "sentence1": "Exploitation of injection flaws almost always requires special characters.", "sentence2": "next to numbers, alphabetical letters can be considered to be a safe character set.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3448", "sentence1": "In this section, we describe the ways that students combined user­centered design and programming sophistication to create their digital stories.", "sentence2": "we also found that, perhaps not surprisingly, when not specifically prompted, students tended to view their project from their own perspective, omitting interactive features that could be understood by an outside user.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3449", "sentence1": "When compared to the programming in Logo, this drastically reduces the number of actions a user needs to make to achieve the same level of functionality (compare Figure 2 and 3).", "sentence2": "the SCP appears to provide the ease of use that comes with a non-programmable control-panel with the functionality of a programming environment.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3450", "sentence1": "However, in scenarios an action does not always end the process.", "sentence2": "an action step in the scenario creates a gap in the plan graph relative to the scenario.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3451", "sentence1": "The key difference to [1] is that the local monitors have a predefined, in general not fully connected, communication topology.", "sentence2": "a local monitor can in general not know the truth of all propositional variables.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3452", "sentence1": "These graphs can then be used by subsequent test­ing processes within the GUITAR framework.", "sentence2": "the study focuses on programs written in C or C++, which use GUI libraries such as GTK or Qt.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3453", "sentence1": "We describe both mechanisms with four groups of design axes related to four questions: When (temporal characteristics), What (content characteristics), How (medium characteristics) and Where (spatial characteristics).", "sentence2": "the What group of feedforward contains a subgroup highlighting the potential impact of the feedback mechanism onto the content of the feedforward mechanism.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3454", "sentence1": "For instance LightGuide [27] displays the gesture to execute through visual cues as if on the user’s hand (egocentric). ", "sentence2": "this projection of the gesture does not follow the hand, but is anchored in a specific location in the world coordinates (absolute).", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3455", "sentence1": "Figure 5 shows an example tree representation of a search space with weights assigned to each sub-branches.", "sentence2": "all of them follows general syntax: name of transformation followed by parameters of the transformation as shown below.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3456", "sentence1": "The idea is to identify regions which the compiler may be able to optimize or specialize using information about values or shapes of certain variables at runtime.", "sentence2": "given the cost of runtime specialization, such regions should contain potentially expensive operations such as loops or ar­ray library operations.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3457", "sentence1": "Furthermore, their appearance reminds the stakeholders of the well-known concept of hyperlinks.", "sentence2": "this approach is intuitively comprehensible for stakeholders and integrates seamlessly into their workflow.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3458", "sentence1": "Engineers and designers prefer free-hand drawing and writing in early design and modeling phases [5, 7].", "sentence2": "the AugIR enforces informal and incomplete sketches and avoids strict modeling languages.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3459", "sentence1": "Because of interchanging between different means of transportation, the traveler needs to process a sequence of unimodal steps.", "sentence2": "the phase during the trip of an intermodal travel chain consists of several unimodal sub chains.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3460", "sentence1": "During the trip/on the vehicle: The traveler needs information in order to navigate to the destination, about the right direction and route.", "sentence2": "intermodal trips place exhaustive demands on support for managing disturbances.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3461", "sentence1": "Technically, all our design choices were possible from within the new CMS, and our IT department was willing to help us implement those choices in whatever way they could.", "sentence2": "during the summer, one of the graduate students in the class was hired to continue the migration of the content from the old site to the new one, and was charged to implement as many of the design changes as possible.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3462", "sentence1": "Active databases that use the event-condition-action paradigm have been extended with fuzzy conditions to handle uncertainty [7].", "sentence2": "they focus on uncertain ECA-rules rather than uncertain data, and they do not support specu­lative calculations.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3463", "sentence1": "The software requirements engineer can observe that the MSD RemoteLockDoors, responsible for the actual locking opera­tion, correctly activates the MSD LockingFeedback triggering the turn signal feedback operation.", "sentence2": "he also reveals that the MSD SpeedLockDoors activates this MSD, if he simulates both use cases together.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3464", "sentence1": "However, he also reveals that the MSD SpeedLockDoors activates this MSD, if he simulates both use cases together.", "sentence2": "the coordination behavior requirements specify that the speed locking activates the turn signals during driving, which can cause safety-critical confusion in the traffic.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3465", "sentence1": "Based on the above analysis, even if cpm.a1 is helped, its end time will still be delayed to the 18th day without cpm.a2 helped.", "sentence2": "currently, many software companies provide dedicated solutions for specific fields by designing universal business flows.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3466", "sentence1": "The different kinds of possible impacts also led to reflections on how to summarize and deal with different values and/or ranges of impacts.", "sentence2": "the benefit is that more semantic and other information is added to the model.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3467", "sentence1": "The annual workshop shall form the basis for a long-term collaboration between the community members.", "sentence2": "the first workshop is conducted as an interactive discussion for the elaboration of initial ideas of how to collect evidence on the impact of agile practices.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3468", "sentence1": "Constraint Programming (CP) typically revolves around de­scribing a global system state, which is why always and once constraints are often sufficient.", "sentence2": "object­oriented Programming (OOP) revolves modularization of system state and dynamic adaption.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3469", "sentence1": "These functions are called from proxies without knowing the internals of their specification.", "sentence2": "proceed calls in the external device would be lost, as external devices do not have information about other available contexts.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3470", "sentence1": "But in practice, programmers will not write their creation code using object algebras unless they are expecting to have to extend their code.", "sentence2": "i do not believe that object algebras provide for unanticipated extension.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3471", "sentence1": "Upon first execution, every message to an object resolves to an object-specific AST snippet (message resolution) [15, 18], which is then inserted into the enclosing AST as a replacement for the message.", "sentence2": "message resolution replaces object-independent messages by object-specific ASTs.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3472", "sentence1": "To this end, FlexJava is a language accompanied with an automated static analysis that supports both fine-and coarse-grained approximation and provides formal safety guarantees.", "sentence2": "any computation and data that exclusively affects the annotated variable is safe to approximate.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3473", "sentence1": "OLED screens are suitable for optimizing energy consumption of GUIs in apps, because the power consumed by OLED screens depends on the combination of color levels in the screen's sub-pixels.", "sentence2": "power models of OLED screens estimate the energy by combining individual consumption of the color sub-pixels [20, 41].", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3474", "sentence1": "OLED screens are suitable for optimizing energy consumption of GUIs in apps, because the power consumed by OLED screens depends on the combination of color levels in the screen's sub-pixels.", "sentence2": "also, by evaluating the solutions on five commercial apps, we confirmed that some project managers and developers are ready to account for GEMMA's recommendations in future app releases.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3475", "sentence1": "Indeed, we can see very clearly in the external contributor models (Table 5) that the more forks a project has, the more pull requests are submitted by non-core developers.", "sentence2": "some projects may have introduced CI because they were experiencing high interest; also, they may have introduced CI because they already had a strong quality culture.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3476", "sentence1": "The loops at lines 14 21 and line 22 implement the second and third steps of SPP respectively.", "sentence2": "the loop at line 22 performs absorption.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3477", "sentence1": "Given two EFSMs, their technique first computes a set of elementary changes, then uses a static dependence analysis to find which test covers what changes, and finally selects to run only tests that are non-redundant with respect to the changes.", "sentence2": "our evaluation does not assume any static analysis and measures the change-related loss in quality for the reduced test suite.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3478", "sentence1": "The preceding studies highlight the efforts spent by researchers on developing and testing methods in design higher education.", "sentence2": "scholars such as Dorst [10] mentions the low uptake of methods among practitioners, a notion shared by Andreasen [2].", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3479", "sentence1": "We run it on the Sora SDR platform and demonstrate on a test-bed that it is able to operate in real-time.", "sentence2": "it is easier to understand and has very similar performance to the hand­tuned Sora code.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3480", "sentence1": "In [7], the authors proposed an approach based on Gray-Box method to generate test cases from UML Activity Diagrams.", "sentence2": "they gave no method to automatically generate test cases from their extracted scenarios.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3481", "sentence1": "Additionally, the factory model contains knowledge about the properties of the different components, their interplay, their locations, their maintenance states, etc.", "sentence2": "the factory model represents the current production capability of the factory.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3482", "sentence1": "For all four metrics described below, classes with high values are more likely to be worse in the SPL quality.", "sentence2": "higher values might be an indicative of a smell affecting the SPL design.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3483", "sentence1": "Constants and refinements are files that can often be found in Feature-Oriented Programming (FOP) [6].", "sentence2": "refinements can change the behavior of a constant if a certain feature is included in a product.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3484", "sentence1": "Features can be defined as modules with consistent, welldefined, independent, and combinable functions [4].", "sentence2": "(vii) Does the method have tool support?", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3485", "sentence1": "To ad­dress this form of variability, techniques related to dynamic software product lines and run-time tenant-level customiza­tion have successfully been applied [21, 26, 30].", "sentence2": "saaS offerings are inherently susceptible to many other forms of variability, caused by (i) co-existence of different versions and variants of key components and enabling services, (ii) variations in deployment (e.g.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3486", "sentence1": "As usual in domain engineering, the models, here the evaluation schemes, are pre-thought for the whole product line.", "sentence2": "for the evaluation schemes it is not yet specified how they are computed, this is done in the activity single view evaluation.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3487", "sentence1": "Moreover, for reusing the automatically derived C2FT as final failure analysis the entire process and tool chain would have to fulfill the requirements of the corresponding safety standard.", "sentence2": "by analyzing and specifying the failure propagation in parallel to the domain architecture, safety aspects can be immediately considered in the architecture and the reuse of C2FTs from the domain into the product implementation facilitates the failure analysis of the final system and saves effort.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3488", "sentence1": "Constraint C4 is based on these archiving events, which are provided by two different probes now, PSteel and PSteelDB.", "sentence2": "analyses [2] can be used after evolution to detect potential anomalies such as dead or false-optional features (see Table 1).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3489", "sentence1": "The initial set of Core components had provided support for the four extension types identified in Figure 1, and standardized a set of contracts for how extensions and applications would interact, exchange data, and persist SME logic.", "sentence2": "we note three key limitations of this experience with respect to evaluation of our hypothesis.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3490", "sentence1": "This paper aims to study the AI techniques from the standpoint of their applications in design phase.", "sentence2": "it focuses on techniques developed or that are being developed (under conceptual stage/s) of AI that can be arranged in solving problems associated with design phase of SDLC.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3491", "sentence1": "The process of this methodology includes set of activities namely; data flow diagram where it converts inputs into the necessary outputs, structural decomposition that is beneficial to improve a structural system model and another activity is detailed design description, which gives the detail knowledge about each function [12].", "sentence2": "there are still certain issues, which are given as follows: FI1: A complex system includes various components and levels of subsystems with multiple interconnections that are complex to manage, recognize and predict [13].", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3492", "sentence1": "However, design phase have a tendency to be production delays, incomplete designs, rework, change orders, and have a big impact in all the development stages.", "sentence2": "the usage of AI techniques in order to improve design phase favorably affects the quality of overall software life cycle [4] [5][6].", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3493", "sentence1": "Starting expertise was better in linear algebra at 3.10 and in C programming at 2.93.", "sentence2": "143 completed the course.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3494", "sentence1": "It describes three algo­rithms to model and predict the satisfaction experienced by individuals using a recommender system for groups, which recommends sequences of items.", "sentence2": "the use of curiosity can aid recommendation systems to achieve better prediction rates and also decrease rejection rate if compared to traditional recommendation systems.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3495", "sentence1": "This paper also discussed future directions for our multimedia validation approach.", "sentence2": "it is possible that the visible amount of a media item has a different form.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3496", "sentence1": "The element behavior is identified through its tag name, its HTML attributes and its JS properties.", "sentence2": "we check the properties that can alter the static behavior of an HTML element, e.g., event listeners, and the HTML5 attributes that identify the role of an element, e.g., \"role\".", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3497", "sentence1": "The non-matching column refers to the average number of blocks that: 1) are over-segmented by the MSoS, 2) have no correspondence with any block in the GT or they are not correctly labeled.", "sentence2": "we check the properties that can alter the static behavior of an HTML element, e.g., event listeners, and the HTML5 attributes that identify the role of an element, e.g., 'role'.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3498", "sentence1": "Furthermore, Eurema uses runtime models as first class entities considering R 6.", "sentence2": "collaboration aspects for the derived requirements R 3 , R 4, R 5 , R 7 , and R 8 are missing.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3499", "sentence1": "We propose new concepts that build on those contained in ensemble-based models.", "sentence2": "we introduce the ideas of ensemble nesting, dynamic role cardinalities and ensemble fitness.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3500", "sentence1": "We also expect many de­pendencies from other repositories towards BioC since it is an active package distribution that offers the same quality checks as CRAN , and also because it contains many use­ful datasets.", "sentence2": "in order to carry out our empirical analysis, we need to extract data from each of the targeted R package reposito­ries.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3501", "sentence1": "Plug-in architectures are a practical vehicle for implementing such extensibility.", "sentence2": "in most contexts, plug-in based mechanisms have been considered as implementation level solutions rather than true architectural patterns that shape the designs as a whole.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3502", "sentence1": "The percentage of rated extensions for Magento (49%) is of particular significance, considering that it is the platform with the greatest number of extensions.", "sentence2": "prestaShop falls short in terms of this metric, accounting only for 23% of rated extensions.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3503", "sentence1": "Regarding the main findings of our work, a number of interesting issues arise: Partner shakeouts as a governance tool.", "sentence2": "it appears that three main platforms account for roughly half of the market share: Magento, PrestaShop, and WooCommerce.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3504", "sentence1": "There are many benefits of automating and optimising architectural assets mass production, but these benefits are not yet exploited fully by today's industries of software-intensive systems, as many issues related to the technical aspects of the development have not been entirely solved.", "sentence2": "it may reduce the severity of the technical issues emerging during the construction process, and other risks related to the decision.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3505", "sentence1": "Ideally, separate fields could be presented for each viewpoint and characteristic.", "sentence2": "the user interfaces in a tool restrict the options, so we condense information that suits our criteria in one field only: subtype.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3506", "sentence1": "Thus, having an overview and understand­ing the architecture of a system is critical for supporting a reasonable evolution, in-line with the desired system quali­ties.", "sentence2": "a considerable emphasis (e.g., [18], [14]) has been put on the need to elaborate software architecture descriptions that depict important views of the considered system from various viewpoints and even two international standards have been released to sustain and systematize the documentation effort ([3], [1]).", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3507", "sentence1": "Testing as well as verification at the concrete level is more expensive than on an abstract one, especially if some corresponding corrections within the system are necessary.", "sentence2": "it makes more sense to have more intensive testing at logical level to reduce the overall size of test suite for the next levels as much as possible.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3508", "sentence1": "The priority and role based scheduler in pSync helps to avoid duplicate file delivery.", "sentence2": "this increases the file sync time when a small number of nodes are out-of-sync.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3509", "sentence1": "This study thus created Logo trend visualization , which is a combination of Radviz and Circle parallel coordinate to provide a guideline to design a competitive logo, based on such background.", "sentence2": "likewise, a relevant article in 2014 also showed that MonoCrest took a hint of Monoline while adding the sense of Rhythm.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3510", "sentence1": "It can be seen from the results that as the number of the kernels increases, the communication overhead becomes significant and the throughput decreases.", "sentence2": "persistent threads approach seems that does not follow the same decay of normal CUDA kernels, providing increased throughput when more than 12 kernels are utilized.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3511", "sentence1": "Symbiosis allows for resolving the same issues by relying on devices that already exist on the network.", "sentence2": "instead of needing more servers, symbiosis allows for perceiving every new purchased smart device as a potential server.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3512", "sentence1": "While all the components presented in Figure 3 are essential, we investigate specific components of the system with a focus on demonstrating SymbIoT’s feasibility, value, and its ability to satisfy its design goals. ", "sentence2": "we describe our initial take on task configuration and generation, nodes configuration advertisement, and scheduling procedures. ", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3513", "sentence1": "Again, the MPI implementation could be made aware of TPSL mappings and make threads share buffers the same way as in a multithreaded implementation, however this would clearly require additional modifications.", "sentence2": "with the considerably increased complexity of recent HPC applications, e.g., demand for components such as in-situ analysis, elaborate monitoring and performance tools, etc., the full availability of Linux APIs has gained high importance.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3514", "sentence1": "The advantage is that the intercept is measured for each clock model separately.", "sentence2": "the intercept error only depends on the accuracy of a single SKaMPI synchronization and on the error of the slope, which was found to be very small (10-8).", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3515", "sentence1": "To compare the synchronization schemes of SKaMPI and Netgauge (NBCBench) to the competitors, we have extracted the relevant clock synchronization algorithms from their respective benchmarking frameworks.", "sentence2": "we use a fixed window size and disable the dynamic window adaptation as done by SKaMPI and NBCBench.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3516", "sentence1": "We show results obtained with MVAPICH 2.0a-qlc, even though it is not the latest version of MVAPICH, but the one that was preinstalled on the system and for which we experienced this significant process skew.", "sentence2": "these results are not meant to evaluate the performance of MVAPICH, but rather to point out potential pitfalls when relying on MPI_Barrier for synchronization.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3517", "sentence1": "SBMA uses interposition to intercept calls to libc's standard malloc library.", "sentence2": "a call to malloc() in a BDMPI program will be performed by SBMA.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3518", "sentence1": "Surprisingly, the performance improvement over the serial application, also in Table 3 was more modest.", "sentence2": "as was discussed in [9], the serial version, by its very nature has a few advantages over a parallel implementation.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3519", "sentence1": "We also note that we use the terms Host and Device interchangeably for CPU and GPU respectively.", "sentence2": "gPU based applications have largely been designed for a single compute node with one or two GPUs.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3520", "sentence1": "The following interpreter gives the same value i on each request: It i a is indeed interpreted as the Reader monad.", "sentence2": "we had to change the data type name and hence modify (the signatures of) addGet and addN, even if their code does not care about the new writer effect and remains essentially the same.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3521", "sentence1": "To manage complexity, information systems and control systems are decomposed in modular building blocks called components.", "sentence2": "the intelligent interface between the information systems and the control systems are also decomposed to fine grained smart interfaces between the components of the information systems and the components of the control systems.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3522", "sentence1": "Language models can only provide help in source code that is similar to source code that it has seen before.", "sentence2": "the data sparsity problem, i.e., the need to see instances of many code con-texts, drives the use of larger and larger corpora to train the model.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3523", "sentence1": "But the reuse of contents is not always possible if it is location-dependent (as is presented in [13]).", "sentence2": "if the content mentions any characteristic of the environment, the reuse is limited.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3524", "sentence1": "We have presented a first version of our tool, which creates location-aware applications in-situ.", "sentence2": "the authors present six dramas with different purposes, for example, tourism and learning.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3525", "sentence1": "However, recall worsens at higher granularities.", "sentence2": "increasing the resolution of analysis decreases the ability of the algorithm to observe drifts occurring at finer granularities.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3526", "sentence1": "Currently, cloud computing systems such as Amazon Web Services (AWS) [3] and Google App Engine (GAE) [16] offer reliability SLAs specifying the fraction of availability over a fixed time period for their services that users can ex­pect, when they contract to use a service.", "sentence2": "they do not provide SLAs guaranteeing minimum levels of perfor­mance.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3527", "sentence1": "The reason why AppScale s performance is more stable over time is because it is deployed on a set of closely con­trolled and monitored cluster of virtual machines (VMs) that use a private Infrastructure-as-a-Service (IaaS) cloud to im­plement isolation.", "sentence2": "the VMs assigned to App-Scale do not share nodes with noisy neighbors in our test environment.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3528", "sentence1": "We also implemented a HDFS client library with HDFS's Namenode as its metadata server.", "sentence2": "since the performance of a single Namenode turns out to be inferior to a single IndexFS server, we only show ShardFS on top of IndexFS servers in the evaluation section.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3529", "sentence1": "Our prototype proved its scalability and in addition can be configured by an experienced user to control the trade-off between monetary cost and number of latency violations.", "sentence2": "this step is optional, in contrast to the mandatory configuration of the user­ defined threshold-based approach.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3530", "sentence1": " SpotOn generally reevaluates its decision whenever a job’s state changes, e.g., due to a revocation, in order to select a new instance type and market to migrate the job.", "sentence2": "on revocation, SpotOn re-executes the greedy cost-aware policy to determine where to restore the job and the fault-tolerance mechanism to use based on the job’s remaining running time.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3531", "sentence1": "So, for comparison purposes, we implemented a working version into PowerGraph and LFGraph unfortunately, this option gave much lower accuracy than Zorro.", "sentence2": "the inaccuracy was 25% for PageRank on PowerGraph and 51% on LFGraph with this option, vs. 0% inaccuracy using Zorro.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3532", "sentence1": "Magpie [3] uses detailed knowledge of application semantics provided by the developer to extract causality from system event logs.", "sentence2": "domino automatically captures causality.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3533", "sentence1": "The reason is that DOM2 callbacks are not enumerable, i.e., given a particular DOM node, Domino cannot retrieve a list of the node's DOM2 handlers.", "sentence2": "domino must interpose on addEventListener(), as we described earlier in this section.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3534", "sentence1": "As shown in Figure 2, Domino interposes on timer registration functions using JavaScript's powerful reflection mechanisms.", "sentence2": "these systems cannot optimize server-side scheduling using knowledge of the \"wide-area\" event stream that spans both clients and servers (§3).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3535", "sentence1": "Since the server-facing portion of the application lacks a GUI, Node does not expose DOM0 or DOM2 events.", "sentence2": "node does support timers and Web Worker-like execution contexts called processes.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3536", "sentence1": "Next, dEst uses the weights of the linear regression from dProf to forecast upcoming cost and benefit.", "sentence2": "we do not have exact information about future objects since these are dependent upon the user's upcoming actions which are obviously not known.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3537", "sentence1": "(2) Non-work-conserving IO schedulers become less effective in virtualized cloud envi­ronments.", "sentence2": "the IO latency under virtualization is much larger than that in native.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3538", "sentence1": "We observe that when three VMs share the same storage, the throughput of VM1 (the low IO-concurrency VM with 16 KB size) is only 60 IOPS using CFQ(Linux) and 1224 IOPS using PS(mClock)  far from the target share 2714 IOPS.", "sentence2": "using BAPS(vFair), the average throughput of VM1 reaches 2721, which is quite close to the target share.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3539", "sentence1": "Typically, it is more efficient to read fewer large records than many small ones.", "sentence2": "a large IO request is slower than a small one due to the device’s data transfer rate.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3540", "sentence1": "Holding a reference in two SingleTLOBs and two space maps during a GC phase nearly doubles the costs (additional 84 bytes).", "sentence2": "the majority of objects are stored in MultiTLOBs, where the costs for maintaining a TLOB can be neglected.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3541", "sentence1": "In order to represent an accurate snapshot of the heap, the trace processing tool needs to merge the object’s characteristics from the allocation event with its new address in the GC move event. ", "sentence2": "the parsed information of the allocation event must be preserved until the associated object dies.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3542", "sentence1": "Control-flow, operation, and collection profiling add a low overhead in ZipPy. ", "sentence2": "variable access and type distribution profiling add a higher overhead.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3543", "sentence1": "There are prior efforts that try to reduce resource consumption on smartphones [38, 44, 39].", "sentence2": "these efforts are orthogonal to our work.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3544", "sentence1": "Tracking apps permission access patterns: The permission access patterns of an app are not exposed to any user-level tool (or even a system-level tool unless it is enforcing the permission).", "sentence2": "zapDroid requires modifications to components of Android to track the permissions accessed by zombie apps.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3545", "sentence1": "No free lunch: By quarantining an app, a user essentially removes the app from her phone (albeit temporarily).", "sentence2": "if she stopped using the app because of the lack of some features that are later incorporated by an update, she must get to know of these new features from an external channel (e.g., a website), since her phone will not receive updates for quarantined apps.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3546", "sentence1": "Broad compatibility: ZapDroid should be usable on most Android devices, if not all.", "sentence2": " it must not be limited to working on a specific vendor’s device(s) or on a particular version of Android.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3547", "sentence1": "Using blockmap, a programmer can express block and cyclic data distribution [28] across MPI processes, while the debugger uses it to construct a global view of the data at runtime.", "sentence2": "languages that manage the decomposition process make it easier for the programmer both during data construction and debugging.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3548", "sentence1": "To simplify validating the ported program, and to help a programmer find the root cause of any errors, relative debugging was proposed in the mid 1990s.", "sentence2": "this ad-hoc approach is impractical in large-scale programs that decompose the data across a deep memory hierarchy for a number of reasons.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3549", "sentence1": "Both the hash-based and P2P methods parallelize data comparison and reduce data traffic between the client and servers.", "sentence2": "a hybrid supercomputer containing heterogeneous processors and a deep memory hierarchy allows programmers to exploit multiple levels of parallelism.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3550", "sentence1": "A few existing studies have also tried to estimate the reliability of a storage system through simulation [7, 8, 11].", "sentence2": "elerath and Pecht [7] have implemented a Monte Carlo simulation for RAID 4 groups to evaluate how time dependent failure and repair rates impact the average number of data loss events that could occur during a given mission time.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3551", "sentence1": "This is mainly because disks constitute only 15-20% of the cost of one SSU.", "sentence2": "when designing a storage system with performance as the primary objective, it is optimal to buy as many SSUs as possible before optimizing or negotiating for disk price or capacity.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3552", "sentence1": "Below 1K nodes, our optimized reorthogonaliza­tion algorithms are not differentiated.", "sentence2": "at extreme scale, this approach is 2-3× faster.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3553", "sentence1": "While it would be possible to manually allocate intermediate value storage, the DSL abstraction allows STELLA to provide backend specific stencil buffer implementations.", "sentence2": "each backend optimizes storage layout, alignment, and memory footprint of the stencil buffers.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3554", "sentence1": "In addition, the reference order of message receive order must be consistent between a record and a replay.", "sentence2": "in the permutation encoding, CDC defines a reference logical-clock order based on Lamport clocks (reference order ), and computes the permutation difference to an actual receive order (observed order )byusing anedit distance algorithm (Section 4.1).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3555", "sentence1": "What is more important is for the application to maintain scalability under record and replay, which CDC enables.", "sentence2": "cDC can replay piggyback clock sends.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3556", "sentence1": "In other words, we posit that programming model research on in-situ analytics is needed.", "sentence2": "in-situ algorithms are currently implemented with low-level parallel programming libraries such as MPI, OpenMP, and Pthread, which offer high performance but require that programmers manually handle all the parallelization complexities.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3557", "sentence1": "For example, consider the graphical depiction of the cumulative sum component in Figure 1, top left.", "sentence2": "such loops are at odds with the anonymity of function results in the lambda-style notation.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3558", "sentence1": "The metadata itself is commonly DRM-specific.", "sentence2": "the TV Middleware Application triggers the DASH Media Player to generate this request.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3559", "sentence1": "The unencrypted content is not exposed over this interface.", "sentence2": "the platform specific information that is exchanged in the browser context in EME is now exposed to the network.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3560", "sentence1": "We presented an application of the analysis contracts methodology to representative analyses from domains of reliability, sensor security, and control.", "sentence2": "a major designtime technique to achieve higher fault-tolerance is redundancy adding functionally identical components in order to preserve system's operation in case one of the components fails.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3561", "sentence1": "Conversely, control safety analysis [20] typically consid­ ers data trustworthiness in the normal operation mode, but often ignores trustworthiness in the failure modes that are provided by FMEA.", "sentence2": "both of these forms of analysis make assumptions that may be inconsistent with, or ignore modes determined by, the other.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3562", "sentence1": "Also pre-escaping and post-escaping strings are completely application and context-sensitive.", "sentence2": "these escaping characters will be determined considering the context and the surrounding characters of the injection point.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3563", "sentence1": "The reason for restricting our case selection to Java projects was a limitation of the used tools for identifying extract method opportunities (see Section 4.3).", "sentence2": "the tool that we used for identifying the extract method opportunities is able of parsing only Eclipse projects.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3564", "sentence1": "On the completion of data collection, a pre-processing step took place.", "sentence2": "we filtered out of the dataset methods that were less prone to suffer from the long method bad smell.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3565", "sentence1": "Until now, the identification of long methods or extract method opportunities has been performed based on cohesion, size or complexity metrics.", "sentence2": "the empirical validation of these metrics has exhibited relatively low accuracy with regard to their capacity to indicate the existence of long methods or extract method opportunities.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3566", "sentence1": "By definition, the long method smell concerns methods large in size, which have a semantic dis­tance between the major purpose of the method, with respect to a specific functionality, and the degree to which its implementation serves this purpose [18].", "sentence2": "we do not perceive all methods large in size as long , but only those whose large size is due to the implementation of multiple functionalities.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3567", "sentence1": "It may seem obvious to inherit techniques and strategies used in general settings.", "sentence2": "there are a number of factors to consider.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3568", "sentence1": "Its economic strength attracts crossborder commuters who represent approximately half of the work force in Luxembourg.", "sentence2": "significant flows of vehicles coming from France, Germany and Belgium significantly contribute to the daily traffic load distribution of Luxembourg. ", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3569", "sentence1": "Additional computations have been done in order to retrieve them.", "sentence2": "according to surveys, France is the most popu­lar holiday destination of the Luxembourgish population, so this matches the observations.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3570", "sentence1": " As Fig. 4 demonstrates, the slowdown forms almost a straight line.", "sentence2": "we build a linear regression model where x is the input data size in GB and y is the total execution time.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3571", "sentence1": "Preliminary observation of this pilot study indicates that users invest for the personalization differently regarding the data type and interaction complexity.", "sentence2": "all participants had an overall positive attitude toward personalizing the interface according to their preferences during use without interrupting their current tasks.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3572", "sentence1": "In some cases one widget totally dominates a specific data type (last name, category, maximum).", "sentence2": "in case of complex interactions more factors need to be considered for instance number of options, multiple/simple choices.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3573", "sentence1": "Our solution relies on the existing trajectory points, obtained from the trace itself, that are organized into clusters to represent anchor points used in the calibration.", "sentence2": "our approach is flexible to be adopted in different real traces, since there is no need for looking at a map nor any further information.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3574", "sentence1": "Adobe Flash is a popular platform frequently used for creation of advertisements, videos and interactive contents on web pages. ", "sentence2": "flash player is neither easy to install nor popular on devices running popular and widely used platforms like Google Android and iOS.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3575", "sentence1": "Service implementation is outside the scope of this paper and we will exclusively focus on versioning in the context of service interfaces, in particular the structural aspect.", "sentence2": "this paper has focused on a solution revolving round the problem of incompatibility in SORC, from both the provider’s and the consumer’s perspectives.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3576", "sentence1": "We can conclude that all the studied factors (pleasantness, unambiguousness, applicability) a.ect the understandability of the diagram.", "sentence2": "their guidelines suggest, e.g., that different graphics and icons should be used modestly, there should be clarity regarding the meaning of different symbols, and color should be preferred over black and white (but the number of di.er­ent colors should be restricted).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3577", "sentence1": "Namespaces will not be advertised by the routing protocol unless given explicit permission by the DNCS (see Step 4 of Figure 1).", "sentence2": "malicious producers may only advertise namespaces under which they are not authorized to serve if they can subvert the routing process verification check or forge a long-term permission token from the DNCS.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3578", "sentence1": "Although our nonoutsourceable puzzle alone does not prevent the collection of statistical evidence, in Section 8.2, we argued that by combining our puzzle with a multi-tier reward system, we effectively make it highly costly or unreliable to accumulate statistical evidence over time. ", "sentence2": "a worker can opt to steal only the “jackpot prize” (which happens only infrequently but offers a large reward), while behaving honestly when it finds a “consolation prize” which is of much smaller amount but paid off at a frequent interval.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3579", "sentence1": "Informally, to effectively outsource work to the worker, the worker must know more than a constant fraction (say, 1/3) of the leaves before calling the random oracle to determine whether an attempt is successful.", "sentence2": "if the worker knows more than 1/3 fraction of the leaves, due to a simple Chernoff bound, it will be able to easily steal the solution should one be found.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3580", "sentence1": "Note that the ID here is bound to the app vendor's server: only this app server is allowed to push messages to the app.", "sentence2": "the study does not go beyond this on the push client side.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3581", "sentence1": "Authentication here relies on Android's security settings: the receiver should be either private or protected by permission, and all the IPC calls should target a specific package.", "sentence2": "he still cannot inject messages to the target app when it verifies user ID.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3582", "sentence1": "Taint propagation algorithms generally propagate taint information at the byte-or word-level, i.e., maintain a taint bit for each byte or word of data.", "sentence2": "the programs in the second set were chosen as representative samples of trigger-based behavior in malware.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3583", "sentence1": "While we have demonstrated promising results, we do not claim that our system is fully mature and has ad­dressed all the challenges.", "sentence2": "we believe that we have made a solid step towards this goal.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3584", "sentence1": "We choose these 100 apps in a mostly random manner but we also consider the dis­tribution of app behaviors.", "sentence2": "40 apps are malware and the others are benign.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3585", "sentence1": "Studies [26, 28] have revealed that the existing descriptions deviate considerably from requested permissions.", "sentence2": "developerdriven description generation cannot be considered trustworthy.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3586", "sentence1": "Furthermore, RPython with PyPy and Truf.e with Truf.e/JS implement complex widely used languages with the goal to optimize the peak performance as much as possible, and indeed reach the performance levels of dedicated JIT compiling VMs.", "sentence2": "meta-tracing has been successfully applied to AST interpreters as well [5], thus, we compare both approaches based on self-optimizing AST interpreters.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3587", "sentence1": "Generally, they advise to expose runtime constants also on the level of the used data structures.", "sentence2": "to prefer fixed sized arrays over variable sized lists, and to use known techniques such as maps [11] to optimize objects to provide the tracer and subsequent optimization with as much information about runtime constants as possible.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3588", "sentence1": "Evaluating each rule at most once: this is accomplished by leveraging rule evaluation order.", "sentence2": "the function applicableRules is designed so that it returns a list of rules.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3589", "sentence1": "This is related to the work on parametric polymorphism and gradual typing [3, 28], which addresses the issue of casting an unknown function type to a parametrically polymorphic type.", "sentence2": "matthews and Ahmed [28] use runtime sealing to ensure that arguments of a polymorphic type variable are used parametrically.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3590", "sentence1": "Cantor s mapping can generalize decomposition into tuples of greater arity (n > 2), but the decomposition becomes inef­.cient to compute [47].", "sentence2": "szudzik s mapping is a bijective function amenable for efficient generalization [46]: .", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3591", "sentence1": "In addition to the synchronization between a VM and its I/O thread, data movement between 2 co-located VMs requires the I/O threads of both VMs to synchronize as well.", "sentence2": "we would need 4 free cores to allow 2 I/O VMs to communicate with each other unimpededly.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3592", "sentence1": "Here, we measure the performance of exporting data from Hive to MySQL.", "sentence2": "hDFS mostly operates in append-only mode, hence we do not need to refresh all information for the mount point.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3593", "sentence1": "Similarly, if a connection consumes the entire bandwidth of a network adapter, without further measures, additional adapters or connections usually do not increase performance.", "sentence2": "usually, the implementation of the protocol is decomposed into functional modules such as the management of client communication, replica message exchange, or the protocol logic.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3594", "sentence1": "Eventually, such an approach leads to a replica architecture where basic tasks are carried out by dedicated threads that exchange their performed work via in-memory queues to preserve the established request order.", "sentence2": "it leads to a task-oriented parallelization [27] and an architecture resembling staged event-driven designs [31].", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3595", "sentence1": " The analysis done so far does not consider the network latency.", "sentence2": "this must be taken into account to avoid that some processes locally stabilize and deliver events be­fore having received concurrent events from other processes.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3596", "sentence1": "For all these extensions, the In­tegrity, Validity, and Total Order properties are not affected, and the proofs of the corresponding results with global time and synchronous rounds can be applied without modification.", "sentence2": "in all cases, the Probabilistic Agreement property needs to be revisited and analyzed more carefully.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3597", "sentence1": "In this paper, we analyze the requirements and the usefulness of the pub/sub paradigm applied to graphs and present GraPS , a graph-based pub/sub middleware.", "sentence2": "the contributions of the paper are as follows: 1. We propose a flexible and powerful graph-based publish/subscribe language that allows users to express interest in parts of a graph.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3598", "sentence1": "In our case studies, we often observe that the set of nodes in the resubscription significantly overlaps with the node set of the original subscription.", "sentence2": "a resubscription operator guarantees the continuity of not missing any publications on nodes in this intersection.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3599", "sentence1": "It shows that our system was able to handle the large amount of publications without stressing the computational resources of the brokers.", "sentence2": "each row of the table contains the subscription identifier as key, the list of node identifiers covered by the subscription, the original query string in case of a graph subscription, and the client identifier of the client that submitted the subscription.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3600", "sentence1": "Both algorithms identify the cycle after 51 simulation steps.", "sentence2": "our approach uses less CDMs through the cycle detection process.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3601", "sentence1": "Cycle detection proceeds as the CDMGraph is being constructed, i.e., with each CDM sent to a process, combined with its snapshot and, after update, sent to another process.", "sentence2": "a CDM carries a consistent view of the fraction of the CDM-Graph already traversed by it, i.e., the processes the CDM has been sent from.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3602", "sentence1": "The Internet Engineering Task Force (IETF) is making a big effort for the standardization of protocols that are aware of energy and processing limitations within the smart objects.", "sentence2": "the IETF working group for Routing Over Low-power and Lossy networks (ROLL) focuses on the design of an IPv6 Routing Protocol for Low power and lossy networks (RPL) [21].", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3603", "sentence1": "To understand the impact of this instability on the per­formance of RPL, we focus on performance indicators such as the number of parent changes and the number of packets lost.", "sentence2": "the former indicates the stability of the topology.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3604", "sentence1": "Once the music mood-mapping of a particular song is performed on a smartphone, it is updated in the centralized cloud database.", "sentence2": "redundant computation of music mood for the same song across multiple smartphones is eliminated.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3605", "sentence1": "It is more skewed for raw configs, because most raw configs are updated by automation tools, which are counted as a single author.", "sentence2": "some configs are updated by a large number of co-authors.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3606", "sentence1": "It minimizes the use of code branches that complicate maintenance.", "sentence2": "frequent software releases increase the risk of software bugs breaking the site.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3607", "sentence1": "One downside of CHL is that all procedures (including internal layers inside the file system) must have explicit pre- and postconditions.", "sentence2": "cHL's approach has two advantages.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3608", "sentence1": "IR view changes require a leader because polling inconsistent replicas can lead to conflicting sets of operations and consensus results.", "sentence2": "the leader must decide on a master record that replicas can then use to synchronize with each other.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3609", "sentence1": "Listing 2 shows the actual code for the list add() function that uses RLU.", "sentence2": "t2 starts a critical section after the call to synchronize rcu() and hence cannot hold a reference to old data.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3610", "sentence1": "The key for correctness is to ensure that RLU protected sections always execute on a consistent memory view (snapshot).", "sentence2": "an RLU protected section must be resistant to possible concurrent overwrites of objects that it currently reads.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3611", "sentence1": "This modification of the global clock using a single operation has the effect of making all the logged changes take effect at the same time.", "sentence2": "if RCU uses a single read-modify-write to switch a pointer to one copy, in RLU the single read-modify-write of the clock switches multiple object copies at once.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3612", "sentence1": "We believe that if one would convert Kyoto to RCU by using the per slot locks for synchronization of writers (like we did), it would provide the same performance as with RLU.", "sentence2": "it is not clear how to even begin to convert those update operations to use RCU.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3613", "sentence1": "Recently Schierl et al. gave an abstract specification of a single file system: UBIFS [32].", "sentence2": "the field oss pid table holds the per-process information tracked by the operating system.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3614", "sentence1": "Traditionally, the storage system had been a SQL database system [48], which is convenient for the developer [34].", "sentence2": "with the emergence of large Web applications in the past decade, the SQL database became a scalability bottleneck.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3615", "sentence1": "Since HTM relies on hardware (CPU) to do concurrency control for local transactions, which is hard to be aborted and rolled back by software.", "sentence2": "to pre­serve serializability among conflicting transactions on mul­tiple nodes, we design a 2PL-like protocol to coordinate accesses to the same database records from local and re­mote worker threads.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3616", "sentence1": "In doing so, P3 may need to dirty the journal and metadata, and will be marked as a proxy for {P1, P2} (the tag is inherited from the page it is writing back).", "sentence2": "p1 and P2 are considered responsible when P3 dirties other pages, and the tag of these pages will be marked as such.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3617", "sentence1": "Note that tasks using direct I/O will not benefit from Duet because they bypass the page cache.", "sentence2": "the maintenance tasks that we have examined do not use direct I/O.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3618", "sentence1": "It does not require tasks to specify their pending work to Duet a-priori.", "sentence2": "a task can change the work it performs while it is running, without informing Duet.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3619", "sentence1": "Our work focuses on long running (large) tasks, whose data may not fit in caches.", "sentence2": "our approach aims to make the best use of the current cached data, by modifying the order in which tasks process data.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3620", "sentence1": "Note that we target real graphs with very high level of node heterogeneity, e.g. small-world, power­law or highly clustered graphs, which leads to small candidate size in most cases.", "sentence2": "the computational complexity of our algorithm is low in real graphs.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3621", "sentence1": "Row 3 shows how AuDroid resolves the secrecy violations relative to the market apps by using the user approval mechanism.", "sentence2": "whenever a market app uses the mi­crophone, the device owner is notified and can approve or deny the access.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3622", "sentence1": "Currently, such operating systems only control access to the microphone, allowing authorized apps to access the microphone at any time, even when a privileged app may be using the speaker or the user may be speaking a sensitive pass-code to the device.", "sentence2": "current mobile operating systems fail in enforcing information flow control through the microphone and speaker, resulting in putting the users data confidentiality and device integrity at risk.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3623", "sentence1": "Furthermore, it may be useful to explore their expectations of how the world should operate and how this might affect their ICT use [63], as it has been implied that at least some differences exist between the younger and older generations of users [31, 64].", "sentence2": "in the current paper we examine how young people are engaging with the Internet and what that implies for ICT design for youth.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3624", "sentence1": "Although in our data the youth were not very eager content creators, related activities were still observable.", "sentence2": "the existing research indicates content creation as an essential aspect to be offered to the youth.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3625", "sentence1": "The kernel already checks the text section for errors with a checksum before booting.", "sentence2": "we assume that read-only code is covered by other mechanisms, such as periodic error checking [30].", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3626", "sentence1": "First, at line 4, Dcurr is not set to the DOM snapshot; instead, Dcurr is built incrementally by adding XPaths to it for the referenced UI elements, one at a time.", "sentence2": "dcurr corresponds to the DOM induced by the test steps in the current candidate subroutine.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3627", "sentence1": "For instance, many test methods may be added, each tagged with the TestMethod attribute, so that the test framework can automatically discover them.", "sentence2": "in summary, the reviews we chose have: between 2 and 5 non-trivial partitions (median 2), between 2 and 13 changed files (median 5), between 7 and 52 diff-regions (median 25), and between 1 and 8 trivial partitions (median 2).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3628", "sentence1": "It is important to note that analyzing changesets is quite different than analyzing the history of a source repository.", "sentence2": "changesets do not provide full information: we do not have the full set of project files, nor the compiler options used to actually compile the code.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3629", "sentence1": "The control and treatment groups had similar average years of programming experience (7.2 years and 7.4 years, respectively), but the control group had overall more experience with VCS (5.8 years) than the treatment group (3.8 years).", "sentence2": "we also tested these hypotheses using two other analyses, to serve as checks.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3630", "sentence1": "Since these edits are likely to have been performed close together in time, users may be able to easily select them together in the timeline.", "sentence2": "if not, we believe the compile errors would catch these kinds of problems and users would be able to perform the remaining backtracking steps to complete what they wanted to achieve.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3631", "sentence1": "It would be much more convenient for her if there was at least a semiautomatic way of restoring the desired code from Fig. 3b while keeping the subsequent desired edits from Fig. 3d. ", "sentence2": "if not, we believe the compile errors would catch these kinds of problems and users would be able to perform the remaining backtracking steps to complete what they wanted to achieve. ", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3632", "sentence1": "Existing qualifier inference tools operate in batch mode.", "sentence2": "they take the source code as input, analyze it, and insert all the qualifiers at once.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3633", "sentence1": "Qualifiers were a new feature of Java by the time of our study.", "sentence2": "we trained the participants about the qualifiers.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3634", "sentence1": "As shown in Figure 6, RECONTEST provides an advantage with long execution traces where, due to the interleaving explosion problem, concurrency testing becomes an expensive activity.", "sentence2": "it is inadequate for concurrent programs to only cover the changes of the sequential logic and forego their ramifications in the interleaving space.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3635", "sentence1": "Most PTAs also re-execute the program to exercise the retained interleavings.", "sentence2": " RECONTEST efficiently characterizes the delta of interleaving space at memory access level.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3636", "sentence1": "As we described in Section III-B3, we distinguish four kinds of program constructs: classes, methods, variables, and com­ments.", "sentence2": "to exploit all of this structural information, we perform a separate search for each type of terms.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3637", "sentence1": "Fortunately, variability-aware refactoring has a mostly local effect on source code.", "sentence2": "in our subject systems, RENAMING IDENTIFIER, EXTRACT FUNCTION, and INLINE FUNCTION usually affected only a few configuration options, which enabled exhaustive testing of all variants.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3638", "sentence1": "The extraction process is implemented as a breadth-first search (BFS) variant, where all the recorded (visited) methods will be skipped.", "sentence2": "all the logging methods are excluded in this process.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3639", "sentence1": "The weekly status meetings provided a lot of information about the developers and the product.", "sentence2": "they provided only the perspective of the meeting participants.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3640", "sentence1": "When performing checks that require small bounds, such as the satisfiability checks, bvzot is only occasionally more efficient than the other tools.", "sentence2": "as the models and bounds5 grow in size, bvzot demonstrates its strengths.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3641", "sentence1": "However, due to undecidability of FOL, an automated theorem prover is not guaranteed to come up with a solution every time, and sometimes it may time-out without providing a conclusive result.", "sentence2": "actions that have loops in them are hardest to check automatically.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3642", "sentence1": "This is similar to our approach at a high level.", "sentence2": "we do static analysis of a particular condition that allows us to use a completely alternate definition of a loop, whereas [8] iteratively abstracts and subsequently reduces the model in order to infer and enhance atomicity rules without altering the validity of the given invariants.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3643", "sentence1": "Users not as engaged with the method will have less motivation to successfully implement an assert, and open source developers appear to follow this trend.", "sentence2": "one issue to note is that many methods have been changed only by a single developer, and therefore these methods have complete ownership (= 1.0).", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3644", "sentence1": "In 91 cases this was true, and in none of the observed error cases was the estimate extremely different from the actual size.", "sentence2": "this estimate is an appropriate measure for roughly distinguishing between different sized functions.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3645", "sentence1": "At the BB level, PRES reproduces all of the bugs within 10 tries.", "sentence2": "in this section, we propose our solutions for the challenges mentioned in Section II.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3646", "sentence1": "Via observing the results, we found that the n-gram model tends to collect APIs including project-specific ones (noises) due to the strict order of n-grams.", "sentence2": "its suggestion accuracy is affected more by noises.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3647", "sentence1": "The API elements in a usage do not always have a specific required order, e.g., the instantiations of the Scanner and FileWriter objects.", "sentence2": "n-gram requires a total order among APIs.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3648", "sentence1": "In the example, the number of loop iterations is controlled by the value of register edi that holds symbolic data (instructions at 0x6116174c and 0x6116174d).", "sentence2": "the value of eax in crash instruction is indirectly controlled by the symbolic input data in edi.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3649", "sentence1": "CARAMEL could have easily followed developers style and generated the same fixes, if WALA was able to provide the line number of the RI.", "sentence2": "in practice, CARAMEL focuses only on RIs that (1) have a single output location, (2) either write a constant (similar to Figure 4(a)) or perform the &#38;= or |= operations (similar to Figure 4(b)), which effectively correspond to's being constants false or true, respectively, and (3) have an output location that is not written to in the loop with other values except S. ", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3650", "sentence1": "Fourth, fixing performance bugs, which manifest for some inputs, may slow down other code, for some other inputs, and developers must decide which of these slowdowns the slowdown caused by the performance bug for some inputs, or the slowdown caused by the fix for some other inputs is preferable.", "sentence2": "fixing performance bugs has benefits, i.e., it speeds up code.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3651", "sentence1": "Imagine the situation where m tests fail due to a certain fault f.", "sentence2": "even if pfii contains only a single potential fault indicator, pfii established this way may include redundant statements that are executed by ti but not all failing test cases that have executed the corresponding fault of pfii.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3652", "sentence1": "In practice, deriving a complete fault indicator for a certain fault can also be difficult.", "sentence2": "note any subset of the state­ments included in the complete fault indicator for a certain fault f can be used to indicate the execution of f for failing tests to a large extent.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3653", "sentence1": "Phase 1: Identifying the suspiciousness score for each failing test's opinion on the potential fault indicator.", "sentence2": "newtestset contains a subset of Ts such that each of them has executed rfi or has not executed any real fault indicators yet its opinion on the potential fault indicator includes a child of rfi.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3654", "sentence1": "Code review usefulness is negatively correlated with the size of a code review.", "sentence2": "the more files there are in a single review, the lower the overall rate of useful feedback.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3655", "sentence1": "In addition, CQ is a general unit which can be reused in other products of LGE.", "sentence2": "it is important to detect bugs in CQ if any.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3656", "sentence1": "This illegal transition was made by the two consecutive events auto-cook and function-L at the same main loop itera­tion on (menu,def ault) state.", "sentence2": "the error occurs when a user presses the auto-cook button then immediately turns the function dial counter-clockwise when the oven is in the menu mode.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3657", "sentence1": "In general, the data shown in Fig. 8 supports the observation that association rules have a relative short lifetime (as suspected in Section VI.B).", "sentence2": "the number of required association rules is high, indicating that false test alarms can manifest in many different ways, but still be consistent over time.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3658", "sentence1": "Please note that a failing test step may not cause the test case to terminate immediately.", "sentence2": "each executed test case may report more than one test step failure each of which may relate to a code defect or a false test alarm.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3659", "sentence1": " Besides precision and recall, we were interested in the number of association rules required to achieve the high precision as discussed in the previous section.", "sentence2": "we wanted to know whether the number of association rules is rather constant—no new rules must be learned over time—or whether constant learning of new rules is required. ", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3660", "sentence1": "While they have historically used metrics related to the overall failure numbers in this analysis, the classification models shown in this paper indicate that test instability often occurs in blocks of time.", "sentence2": "when attempting to identify troublesome tests, it is useful to look not only at total failures, but also the recency of failures within a given window of time.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3661", "sentence1": "Different classification models yield varying levels of prediction accuracy depending on the data set and the input parameters chosen [16].", "sentence2": "we also need to examine various classification algorithms, such as decision tree models and Bayesian models.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3662", "sentence1": "WFD 2 SPF was also used to provide higher-level communication abstractions for the SPF.", "sentence2": "we introduced three communication primitives: one for sending broadcast messages, one for sending unicast messages, and an RPC-friendly primitive for sending a message and waiting for its response.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3663", "sentence1": "This is an abstract class that contains the logic that is common to group owners and regular group members.", "sentence2": "it contains methods to connect and disconnect, to perform messaging, and to respond to incoming messages.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3664", "sentence1": "The recently committed changes may be buggy but have not been detected and fixed yet.", "sentence2": "we apply our prediction models on the changes of the proprietary software made in the last year.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3665", "sentence1": " We have identified the following main reasons.", "sentence2": "the time period of the test set is often short, e.g., a few days or months.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3666", "sentence1": "If a project's BTS is well maintained and linked, we consider changes whose commit messages contain a bug report ID bug-fixing changes.", "sentence2": "a line that is deleted or changed by a bug-fixing change is a faulty line.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3667", "sentence1": "This limits the possibility for more developers to participate in the field trial.", "sentence2": "the feedback from the developers may not be representative.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3668", "sentence1": "First, in practice, we prefer to predict as soon as changes are committed so that we can examine them to identify bugs earlier.", "sentence2": "the time period of the test set is often short, e.g., a few days or months.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3669", "sentence1": "The overall result of this study is the definition of a metric to aid stakeholders in deciding on whether the dependencies of a system should be updated.", "sentence2": "the contributions of this paper are: The definition of several metrics to quantify dependency freshness at the component-level, including their advantages and disadvantages.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3670", "sentence1": "These criteria are even more essential in a context whereby the safety-critical system under analysis is being developed in an incremental manner and that formal proof is used as a means to ensure non-regression.", "sentence2": "any counterexample generated from a failed proof attempt can also be simulated on the model for debugging purposes.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3671", "sentence1": "These tables show that only 6 root spaces are needed to cover 89% of the Bug2 space, and 7 root spaces can cover 92% of the Change10 space, meaning that both error-prone and change-prone files are highly architecturally connected.", "sentence2": "we observe that the first two DRSpaces are the same for both Bug2 and Change10, and cover up to 76% and 79% of the most error and change prone files respectively.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3672", "sentence1": "RQ3: Is it possible to quantify the return on investment of removing architecture debts?", "sentence2": "is it possible to determine the penalty incurred by the debts and the expected benefits if the debts are removed, and compare this with the costs of refactoring?", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3673", "sentence1": "Since any given file (class) may participate in many relationships with other files, this may result in the same file appearing in multiple DRSpaces.", "sentence2": "we calculated two measures of the size of each DRSpace: the raw size, in terms of number of files, and a normalized size.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3674", "sentence1": "We are not aware of any other tools that can detect those files, together with their visualizable architecture issues.", "sentence2": "this is, to our knowledge, the first time that the penalty associated with technical debt, the cost of refactoring to remove that debt, and the expected benefits of removing the debt have been quantified based on hard data project-specific empirical evidence.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3675", "sentence1": "The goal of this work is to shift developers toward more effective program comprehension and maintenance habits by providing an approach that fosters structural code navigation.", "sentence2": "nC then represents the sum of all structured navigation in Visual Studio and in Prodet.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3676", "sentence1": "Software Engineering educators are expected to provide students with hands-on learning activities that challenge them to solve authentic (i.e., real or very close to real world) problems by applying the relevant Software Engineering concepts and principles.", "sentence2": "there is an increasing realization and focus on problem based and experienced based teaching and learning in Software Engineering education of different phases of Software Engineering such as Requirement Engineering, Design, Implementation, and Testing.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3677", "sentence1": "While not statistically significant, nor surprising, students reported that the Scrum practices were slightly less suited to working in a distributed team than in a local one and that the process conformance was slightly poorer when the teams were distributed.", "sentence2": "in the interviews they stated the opposite: several students mentioned that in distributed teams they were actually following the agreed practices a bit better than in the collocated teams, e.g., communicating right away in Flowdock when working on tasks or having problems, instead of waiting for the next face-to-face meeting, which did happen when working in a local team.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3678", "sentence1": "The PO created user stories with less dependencies over time, as shown in Table VII.", "sentence2": "i think I like the international team, a lot better.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3679", "sentence1": "However, even though they have been taught software engineering (SE) principles (both in SCC.230 and SCC.330), because these have been taught through practice, the students are not always able to explicitly articulate their learning goals in terms of SE principles.", "sentence2": "we more focus on individual attainment of software engineering principles as well as system implementation techniques in SCC.331.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3680", "sentence1": "To cater for this, we propose that we integrate components from the Contextual Design participatory design method [20].", "sentence2": "the Contextual Design method utilises the notion of a Cultural Model that provides an analytical capability for showing the cultural or political forces in the organisation.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3681", "sentence1": "We observe that the patch applied in the onCreate method (Patch2) avoids the crash, whereas the patch applied in the getWorldList method (Patch1) continues throwing the exception.", "sentence2": "the store learns that Patch1 is not effective and automatically discards it.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3682", "sentence1": "We then produce clusters of similar (code, specification) pairs.", "sentence2": "we will use the similarity measures for finding similar code, code with similar usages, and code with same/similar specifications.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3683", "sentence1": "In short, there is a need for a new class of safety certifi­cation/assurance techniques that are continually assessing and evolving the safety reasoning, concurrently with the system, to provide through-life safety assurance.", "sentence2": "safety assurance is provided not only during initial development and deploy­ment, but also at runtime based on operational data.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3684", "sentence1": "Moreover, we found that participants verbally described operations in terms of code fragments, e.g., move this con­ditional statement, yet version control tools required them to think in terms of lines.", "sentence2": "he edits the autogenerated commit message (Figure 1C) to better reflect his intent, by clicking and then typing the new message.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3685", "sentence1": "In the first step, we remove markup from the HTML files and preprocess the resulting text files to account for the unique characteristics of software documentation not found in other texts, such as the systematic use of incomplete sentences and the presence of code terms.", "sentence2": "we prefix sentences that start with a verb in present tense, third person singular, such as returns or computes , with the word this to ensure correct parsing of partial sentences, such as Returns the next page number .", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3686", "sentence1": "Abstract Developers have to work with ever-present designtime uncertainty, i.e., uncertainty about selecting among alternative design decisions.", "sentence2": "existing tools do not support working in the presence of uncertainty, forcing developers to either make provisional, premature decisions, or to avoid using the tools altogether until uncertainty is resolved.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3687", "sentence1": "A detailed discussion of these annotations is beyond the scope of this paper.", "sentence2": "tesMa takes as input a set of design documents of an application under test.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3688", "sentence1": "Multi-version projects are better choices as they have several programs whose sizes increase gradually, making it more obvious to observe the 2In this work, the cost of selective mutation testing refers to the time of compiling and executing every selected mutants against each test case.", "sentence2": "more selected mutants means larger cost.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3689", "sentence1": "Linear relationship may mean that the cost for applying selective mutation testing changes greatly as the program size increases.", "sentence2": "though selective mutation testing does not use all the mutants, the number of selected mutants used in selective mutation testing for large programs may still be large, which is still expensive to run every selected mutants against the test cases.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3690", "sentence1": "At one end of the scale, specified oracles can be generated from formal specifications [3], and are effective in identifying failures, but de.ning and maintaining such specifications is demanding and consequently such specifications are very rare.", "sentence2": "a number of different cluster counts were explored based on a percentage of the number of subject program test cases (1%, 5%, 10%, 15%, 20% and 25%).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3691", "sentence1": "The trend is for the failure density of the clusters to increase in line with the number of clusters (with the exception of the aforementioned version 3 in NanoXML).", "sentence2": "in some cases, as the number of clusters increases the failure intensity peaks and then begins to drop (although not substantially) as the clusters are forced to fragment.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3692", "sentence1": "There has been much research into methods on how to explore and test a GUI [14], [8], [3], [10], [4].", "sentence2": "little research has been done on the effect of test case length in regards to GUI tests.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3693", "sentence1": "As well, it is possible that one fault causes more than one type of exception to be thrown.", "sentence2": "figure 3 shows the EPC for each application.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3694", "sentence1": "As in the AIM-adaptive scenario AIM completely re-instruments the AUT for each experiment, there are no unnecessary (or disabled) probes introducing additional overhead.", "sentence2": "the AIM­adaptive scenario provides the most accurate prediction results with relative prediction errors between 0.01% and 3.5%.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3695", "sentence1": "Therefore, a parallel unit test usually has no assert statements - as the decision whether a parallelization bug exists cannot be cast by an assert statement, but by an external testing tool.", "sentence2": "most generation approaches for parallel unit tests omit any assert statements - conventional bugs may still be found using conventional unit tests.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3696", "sentence1": "The level of obtrusiveness is an appropriate dimension to position different research strategies on, as it indicates the level of involvement of participants in a study or the level of involvement that a researcher has in creating the study setting.", "sentence2": "the term simulation in this context refers to the artificially created setting in which the experiment or the observed behavior takes place.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3697", "sentence1": "Field studies, as they are defined here, are unobtrusive.", "sentence2": "they do not include any deliberate modification of the environment in which the research is conducted.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3698", "sentence1": "The answers the participants provided to the open-ended questions in all eight areas related mostly to the human and collaborative aspects of their activities.", "sentence2": "we included in our analysis of the open questions the responses obtained in the context of all eight areas.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3699", "sentence1": "Results of mediation analysis showed that value congruence was a mediating factor between candidate success factors and project success in this model.", "sentence2": "full mediation existed between transformational leadership and project success, and partial mediation existed between degree of agility and project success.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3700", "sentence1": "LSI requires defining the number of topics that should be extracted.", "sentence2": "the pure Jaccard performed better for comparing tasks of different developers.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3701", "sentence1": "Most tools resort to approximation to overcome these obstacles and it sometimes leads to incorrect results.", "sentence2": "tool effectiveness needs to be evaluated.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3702", "sentence1": "Measure element is a term used in the ISO/IEC 25000 series of system and software product quality standards for describing anti-patterns that can be aggregated to form a measure of a particular quality characteristic such as Reliability, Security, or Maintainability.", "sentence2": "each type of defect, weakness or anti­pattern, can be treated as a measure element included in a measure focused on a specific quality characteristic.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3703", "sentence1": "Several evaluating metrics on software reliability are available and effectively assess the performance of SRGMs from the different profiles of model attributes.", "sentence2": "these metrics only evaluate one profile of model attributes and cannot comprehensively evaluate the complete profile of model attributes.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3704", "sentence1": "Objectively speaking, perfect fault detection models to the extent can effectively assess the software reliability in software testing.", "sentence2": "in DS5, parameters are estimated by using time t = 1 to t = 10 weeks.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3705", "sentence1": "Especially, in modern large software systems, detected complex faults and failures are harder to isolate, reproduce and remove compared with detected simple faults and failures.", "sentence2": "the fault detection rate presents the irregular fluctuation over time in the software testing process.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3706", "sentence1": "Among all models, the proposed model has the lowest fitting and predictive SSE and BIAS values.", "sentence2": "the proposed model has much better performance than other models.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3707", "sentence1": "The Verification and Validation (V&V) of this access control policy is thus a crucial activity in the design of the SIS.", "sentence2": "the access control policy should prevent insiders attacks where legitimate users of the system abuse of their privileges to gain access to forbidden information.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3708", "sentence1": "Specifically, each interaction must be analyzed to determine if it is a problem; if so, then a patch must be designed, implemented, and tested.", "sentence2": "a measure of the number of ways in which features in a product line can interact would tell the developer more about the amount of effort needed to integrate features than a simple interaction-existence check provides.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3709", "sentence1": "This modularization strategy, from both the game-template architecture and the code derivation perspectives, has substantially eased the evolution of the first version of PhyDSLinto PhyDSL-2.", "sentence2": "when we introduced the camera management and on-screen control features into PhyDSL-2, we found that (i) changes were limited to one game-template architecture component, (ii) language changes were progressively made from the impacted intermediate model, to the domain-specific language, thus modifying fewer M2M and M2T transformations, and (iii) developers experienced less severe cognitive challenges while executing maintenance tasks.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3710", "sentence1": "Moreover, the heuristics are based on the assumption that the source code contribution originates from the same person which changes the issue status to resolved.", "sentence2": "this assumption does not always hold and depends on the specific project.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3711", "sentence1": "The reason is that human instructions did not include a clear description of how to show the Advanced panel option in the menu, but the CRASHDROID report included the step (change from the basic to the Hex panel) that activates the option in the menu: I swiped right on the top half of the screen: I touched the Hex button...I touched the Advanced panel menu option.", "sentence2": "the qualitative and quantitative results suggest that CRASHDROID reports are as effective natural language descriptions as those produced by developers in this context.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3712", "sentence1": "They focus on visualizing coverage information, and dependencies between tests to help test engineers understanding relations between test cases.", "sentence2": "they do not focus on clone relationships between tests.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3713", "sentence1": "First, we propose that triggers and actions are the only non­derivative concern roles.", "sentence2": "the other roles, including initializers, data, and connectors, can be derived if the triggers and actions are known, yet triggers and actions cannot (in the general case) be derived from the other roles.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3714", "sentence1": "The developer could set a breakpoint in the code and observe which of the multiple implementations are invoked at run time, but this usually gives just a narrow selection of all possible invocations.", "sentence2": "unless she sees all possible invocations occurring at that particular place in source code, she will not be able to fully understand the behavior of the ba­sicDisplayBox message.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3715", "sentence1": "Makefiles generated using CMake and QMake were easy to identify because they must be generated manually.", "sentence2": "makefiles generated with Automake posed a challenge.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3716", "sentence1": "Other features such as special targets, continuations, and recipe com­mand flags ( @ , - , and + ) are used by most.", "sentence2": "there are two flavours of variables: recursively expanded, which are evaluated as they are needed (i.e., when a reference is read), and simply expanded, which are evaluated when they are defined.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3717", "sentence1": "For the first experiment, each subject had a single session.", "sentence2": "most subjects used the same computer.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3718", "sentence1": "Our subjects were only made up of students.", "sentence2": "related work is discussed in Section V. ", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3719", "sentence1": "Those facts indicated that reorderings by the proposed technique did not make the results of keyword-based code searches worse.", "sentence2": "we consider that reorderings by the proposed technique worked well especially when many keywords are found by the keyword-based code search.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3720", "sentence1": "Seahawk [5] is another Eclipse plugin that integrates Stack-Overflow into the IDE.", "sentence2": "seahawk automatically creates queries from the active context in the IDE, presents a ranked and interactive list of results, lets users import code snippets in discussions through drag & drop.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3721", "sentence1": "Existing approaches to find a solution to an exception mainly suffer from several limitations.", "sentence2": "none of them recommend both code snippets and relevant program­mers discussions together.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3722", "sentence1": "If each design model or code conforms to Archface in 2), Fp is also satisfied in the model and code.", "sentence2": "the verification of Fp results in modular interface checking.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3723", "sentence1": "more than 33 000 changes between two consecutive releases 4.1.3 and 4.2.1 as shown later), careful analysis of the impact of these changes on different modeling tools supporting these solutions is required before their implementation [6].", "sentence2": "these kinds of analysis would not be feasible for the automotive software designers without the tool support.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3724", "sentence1": "The distcheck tool is also being used to schedule package rebuilds within Debian in a way that is aware of unsatisfiable build-time dependencies (as opposed to deploy-time dependencies, which are the main concern of this paper).", "sentence2": "no build will be scheduled by the wanna-build scheduler, unless distcheck can show that the build-time dependencies of a given source package are currently satisfiable in the origin repository.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3725", "sentence1": "They found different coupling patterns in projects with different development style.", "sentence2": "in test-driven projects there is a strong coupling between production and test code, while other projects have a weaker coupling between testing and development.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3726", "sentence1": "Additionally, the destination of these incidents is a Text editor in more than 65% of the incidents and is a Java editor in 11% of the incidents.", "sentence2": "the interaction between the C language and other editor types is minimal because of the nature of the C language itself which is primarily used to develop standalone applications that do not require text to be moved across different editor types.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3727", "sentence1": "The elementary patterns are composed of a single C&P interaction involving one or more files. ", "sentence2": "complex patterns are composed of two or more C&P incidents involving more than two files.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3728", "sentence1": "The technical dependencies that exist between projects define the structure of the ecosystem [2].", "sentence2": "identifying technical dependencies between software projects is a useful way to identify ecosystems.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3729", "sentence1": "At the same time, there is a relative scarcity of empirical data about the nature of architectural change.", "sentence2": "the largest versions of these systems range between 150KSLOC and 800KSLOC.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3730", "sentence1": "On the other hand, in order to meaningfully compare two concern-based architectures as required for our study, we needed a shared topic model for their recovery.", "sentence2": "for each subject system, we created a topic model by using all available versions of the system as the input to MALLET.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3731", "sentence1": "As a result, the architectural gap between version 0.1.0 and 1.0.0 is expected to be very large, yielding a low a2aMa jor value.", "sentence2": "changes between the last minor version and the subsequent major version that is derived from it (i.e., for the version pair (0.20.0,1.0.0)) are comparatively small, resulting in a relatively high a2aMinMa j value.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3732", "sentence1": "In addition, we have attempted to cover most important risk factors related to an issue causing a project delay.", "sentence2": "we acknowledge that the set of risk factors identified in this paper are by no means comprehensive to encompass all aspects of software projects.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3733", "sentence1": "The main reason for using NB is to obtain the probability that a issue will be delayed as we need to determine a likelihood of the risk.", "sentence2": "studies that aim to measure the degree of uncertainty naturally apply NB to build models (e.g., [37, 38]).", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3734", "sentence1": "First, they are trained to predict the best answer from any lists based on certain metrics.", "sentence2": "they are not expected to explain why none of the answers would be chosen as a solution.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3735", "sentence1": "To answer RQ1, we analyze 3,956 unresolved questions from Stack Overflow using four different aspects lexical, semantic, user behaviour and popularity.", "sentence2": "we compare the unresolved questions with resolved questions from data gathering step (Section II) using those aspects, and attempt to determine if there exist any noticeable differences between them.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3736", "sentence1": "Moreover, previous research has shown how the users with a high reputation are more effective in providing successful answers [1],[3].", "sentence2": "we include the social reputation of the answerer as a control factor in our model.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3737", "sentence1": "In the context of ecosystem, the platform provider may want to design its APIs simple enough to encourage niche creation.", "sentence2": "an API providing higher level of penetrability to the platform would be more appropriate to more established partner companies who prefer to have more understanding and control of the platform.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3738", "sentence1": "The goal in this work is to understand whether the meta-model of ABRE-QM enables a definition of quality in this context.", "sentence2": "by extracting this abstract information, we were able to keep impacts on e.g. implementing a use case only if the impact was really specific to this activity and not due to a generalized one of the above.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3739", "sentence1": "Similarly we can split up activities to further understand which activities are affected in what way.", "sentence2": "a pure analysis of the process model description is not sufficient, but a more liberal interpretation is required.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3740", "sentence1": "However, a pure analysis of the process model description is not sufficient, but a more liberal interpretation is required.", "sentence2": "a static (idealized) description of a process is not sufficient as the obtained quality factors (and their practical value) highly depend on the person deriving it, especially regarding the expertise in the field and, since definitions are vague, experiences how the activities are carried out in practice.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3741", "sentence1": "We have described the experience made at SEAC, where we introduced a new, agile development process, targeting several issues of an ongoing re-engineering project.", "sentence2": "we focused on the derivation of the implementation and of the test cases from the normative requirements, so as to ensure alignment between the testing goals and the expected system behaviours expressed in the requirements documents.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3742", "sentence1": "The configuration of LSI, which depends on the underlying dataset, greatly influences the accuracy of the results.", "sentence2": "one of the key challenges in applying LSI-based methods is finding an appropriate configuration producing accurate results.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3743", "sentence1": "How­ever, the work focuses on understanding the inner mechanisms of LSI and does not consider the accuracy of the results.", "sentence2": "the accuracy is in the center of our work, since it improves the applicability of LSI in requirements tracing.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3744", "sentence1": "We believe that addressing all of these requirements is essential for the creation of a robust utility network model which can support the requirements of modern utilities.", "sentence2": "another category of dense features is associated with substations.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3745", "sentence1": "Several new programming environments have been introduced recently, in order to address these issues.", "sentence2": "the OpenACC programming standard has been designed to ease the software development process for codes targeted for heterogeneous machines, helping to achieve code and performance portability.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3746", "sentence1": "It expedites model modification and enhancement; it also enables environmental model reconfiguration, reuse and reassembly.", "sentence2": "they contribute organic carbon throughout the soil profile in the form of labile root exudates and decomposing root material, and strongly influence ecosystem CO2 and CH4fluxes through respiration, rhizosphere oxygenation, and passive transport (e.g. [4]). ", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3747", "sentence1": "Based on the results from the previous step, we have a clear understanding of the software structure and data dependency (inflow and outflow) of any specific scientific function within environmental models.", "sentence2": "for any given scientific function, the information on software structure and data dependency can be used to generate code segments for test module initialization (if necessary), and input/output.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3748", "sentence1": "While GAs may provide acceptable performance, other al­gorithms may be better at exploiting the particular structure of a testing problem: the focus on GAs may be limiting our creativity as researchers in finding much better .ts between the testing problem and the search algorithm.", "sentence2": "gAs can scale poorly to high-dimensional problems and scalability is sensitive to the representation of the problem (see, for example, [4]).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3749", "sentence1": "This has been exemplified in recent work [15].", "sentence2": "it has been shown that it is the combination of soft and hard constraints that allows for a very popular CIT-test generation algorithm, based on simulated­annealing, to scale to real-world problems.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3750", "sentence1": "The results show that the use of EAs is suitable in many cases.", "sentence2": "it can be outperformed by simpler search techniques.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3751", "sentence1": "To automate the test data generation process, we have considered that our System Under Test (SUT) is like a gray/semi-transparent box in where the internal structure is partially known.", "sentence2": "we can design test cases through the exposed interfaces and conduct a code coverage analysis from the general structure of our target SUT.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3752", "sentence1": "This identification is based on the computation of the differences between the FM before the evolution and the evolved one.", "sentence2": "it relies on a syntactic difference diff operator that takes two FMs as input and computes their differences [19].", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3753", "sentence1": "TABLE II describes the edits that can be performed on cross-tree constraints, whether Boolean or non-Boolean ones, e.g., constraints on the number of instances of a given feature [16].", "sentence2": "this architecture is able to handle the evolution of FMs extended with both attributes and cardinalities.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3754", "sentence1": "Also, design-time analysis, as it is presented here, is an exploratory approach compared to run-time analysis.", "sentence2": "the questions that arise while conducting this method would not otherwise come to light if the focus had only been on the run-time analysis of the self-adaptive system.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3755", "sentence1": "In the next section, an application of the method on a self-adaptive system is presented using hypothetical data.", "sentence2": "the quantified NFRs, P(NFRi), P(NFRID), risks as­sociated with each NFR, the outcome attributes, their weights, values, and the percentage effectiveness of each design deci­sion in mitigating the risk are all inspired by empirical data provided by a prototype of the case study.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3756", "sentence1": "Calculating surprise values for the river depths greater than 1.53 m yields a negative surprise value.", "sentence2": "as the magnitude of surprise matters the most, the sign is ignored.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3757", "sentence1": "Perfect knowledge, sound judgement, and a consensus of opinion among the considered stakeholders are not realistic to assume.", "sentence2": "the Software Engineering literature provides pragmatic solutions and systematic approaches that, if adopted, can minimize the effect of this problem.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3758", "sentence1": "For this paper, we use the definitions of test plans and test cases as defined by the IEEE [29].", "sentence2": "a test case comprises an expected value and the conditions necessary for execution.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3759", "sentence1": "Lastly, the total number of executed test cases were recorded to provide a measure of the overall impact of run-time testing to an SAS.", "sentence2": " we demonstrate that Proteus significantly reduces the amount of executed test cases per testing cycle by ensuring that only relevant test cases are executed, as shown in Figure 7.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3760", "sentence1": "Video streaming is an attractive mechanism for technology transfer at a reduced cost (both time and money) per viewer - particularly for companies with multiple locations (including within the same city, campus, or building).", "sentence2": "using video effectively is a challenge when it comes to audience interaction since the medium is much more limiting than a live lecture.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3761", "sentence1": "On the basis of a successful alignment of expectations and pre-study, the industrial PhD project proved in our case to be a good instrument for bringing the technology safely from TRL 5 to TRL 7, that is, through the valley of death.", "sentence2": "there are still challenges and risks: The strong involvement of a company hiring the PhD candidate and typically paying two thirds of the salary gives a risk of the company not being able (or willing to) complete the project.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3762", "sentence1": "This could make industry partners more likely to leave a research project before its completion.", "sentence2": "bridging from TRL level 5, validating the technology in relevant environment, to level 7, the point where industrial partners can convincingly see the developed technologies being applied in practice, requires often much more investment of the industrial partners.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3763", "sentence1": "As a result, many potentially useful research results lie fallow, to be reinvented sometimes by practitioners but, typically, in a more limited and, hence, less useful form that is imposed by the specific technology and problem on hand.", "sentence2": "shunned by industry, many researchers do indeed turn to investigating problems that are of little or no practical relevance, which, nevertheless, allows them to publish and thus maintain or enhance their academic standings.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3764", "sentence1": "The topmost node is a choice between four alternatives: The leftmost subtree checks whether the robot is in one of its home locations.", "sentence2": "finally, we discuss some aspects of the overall process of contin­uous collaboration, a development pattern based on using feedback gained from the running system to continuously adapt and improve a system's operation throughout its life cycle (subsection D).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3765", "sentence1": "Each node created during the development process can be formally described using the formal system model described above.", "sentence2": " the intention of this paper is to propose an extension of the formal system model, which on the one hand allows to capture uncertainty factors that are inherent in the context of a CPS, while on the other hand it should support all modeling concepts of the initial model such as composition and refinement.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3766", "sentence1": "Such a strategy defines the membership function of a property according to the observed history of a channel.", "sentence2": "the property adapts to the location of a component.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3767", "sentence1": "Context-aware systems are characterized by their ability to interact with their external environment by monitoring information over sensors, or by controlling system states over actuators.", "sentence2": "it might be beneficial if the VPP could dynamically reconfigure its specification.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3768", "sentence1": "In this paper, we illustrate the use of the mKAOS language and its associated tool by modeling the missions of a Flood Monitoring SoS.", "sentence2": "we argue that missions require additional information that cannot be represented through traditional requirements approaches, e.g., the interactions among constituent systems that lead to the emergent behaviors of an SoS.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3769", "sentence1": "Evaluation Results We only consider the results of 17 high-level recovery modules, because the extraction of meta data has an impact on high-level recovery modules.", "sentence2": "our ap­proach consists of an active repository that is steadily ana­lyzed to recover traceability information.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3770", "sentence1": "The architecture needs to support this way of working and there are significant architectural implications.", "sentence2": "continuous deployment, split testing and data collection need to be an integral part of the architecture.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3771", "sentence1": "Search Provide support for search functionality Claims Assessor should be able to search for the claims record to be processed.", "sentence2": "the manual identification and classification took approximately 7.5 hours of each person s time.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3772", "sentence1": "For SCJ Level 1, we will mainly test the performance of this protocol as the paper has a clear description of its behavior under partitioned systems.", "sentence2": "the test cases are either Level 1 or Level 2 application programs.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3773", "sentence1": "In addition, a dependency analysis facility is supported by the icecap tool chain to check whether all the referenced Java classes are provided by the SCJ RTE.", "sentence2": "before compiling a Java program, dependency analysis is performed to ensure all the referenced classes are supported, and then only the components being used in the libraries will be included in the compilation and the final executable file.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3774", "sentence1": "A restricted version of no heap real-time threads is also provided in order to support a thread-based concurrency model.", "sentence2": "one disadvantage of this approach is that the new structure has to rely on an underlying RTOS for scheduling purpose, which makes the HVM more difficult to port to a standalone environment.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3775", "sentence1": "The iOS application goes a step further and makes all of the data available for of.ine browsing.", "sentence2": "node b and node c are removed in separate mutations.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3776", "sentence1": "In this paper, we present the design of TouchDevelop’s cloud back-end and how it simplifies the creation, maintenance and publication of apps.", "sentence2": "we believe that for beginning app authors, fully featured version control systems together with app store publication models are too complex.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3777", "sentence1": "In the case of existing back-ends, the model of the to-be-developed application is implicit in the back-end API/resource schema.", "sentence2": "finally, the WL++ code­generation engine is endowed with a number of code templates and external libraries, which it refines based on the application model to produce the application code.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3778", "sentence1": "Since the bring-your-own-device (BYOD) trend is continuing to grow, mobile devices can proliferate in a company at an exponential rate.", "sentence2": "metaXA is calling for an infrastructure that can dynamically and automatically scale up according to the growing number of mobile devices.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3779", "sentence1": "Within the terminology of FDPS, the interaction between particles in the gravitational N-body problem is of the “longrange” type.", "sentence2": "we need to specify the function to calculate interactions for both the ordinary particles and superparticles.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3780", "sentence1": "FDPS pro­vides the former part and the users write the latter.", "sentence2": "in simulations with a large number of particles N, NJ,i and NS,i are many orders of magnitude smaller than N.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3781", "sentence1": "For gravity calculations, we used highly-optimized kernel developed using SIMD builtin functions on both platforms.", "sentence2": "for fluid calculations, we used the optimized kernel on the K computer, but do not on Cray XC30.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3782", "sentence1": "Our code using FDPS seems to be 4 times slower than GreeM.", "sentence2": "this is due to the difference of the algorithms used.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3783", "sentence1": "The user program first uses the functions of FDPS to do the in­teraction calculation, and then updates the physical data of particles, and repeats this loop until the end of simulation.", "sentence2": "the user of FDPS does not have to write complex code for parallelization.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3784", "sentence1": "Likewise, modern build systems only recompile compilation units whose source code has changed. ", "sentence2": "during active development using Puffin, only the files that are modified must be recompiled.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3785", "sentence1": "These numbers have been normal­ized based the Original version of LULESH 2.0.", "sentence2": "a time of 1.08 means that the All Features version ran 8% slower than the Original version of LULESH 2.0 for the test with problem size of 20.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3786", "sentence1": "Liszt supports both CPU- and GPU-based parallelism, but does not support incremental adoption.", "sentence2": "entire applications must be written in Liszt to use Liszt.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3787", "sentence1": "Since all elements in LULESH 2.0 are hexahedrons, each element has eight corners.", "sentence2": "each element is affiliated with eight nodes.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3788", "sentence1": "The reason is that significant pieces of HPC knowledge are indeed not prior knowledge, but highly dependent on a particular execution instance of a program running on a given environment.", "sentence2": "it is impossible to prepare a static, prebuilt, all-you-need knowledge base for HPC.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3789", "sentence1": "We provide results demonstrating significant reduction in GPU overhead, allowing tasks where speedups over the CPU version were previously unattainable to now outperform their CPU counterparts.", "sentence2": "doing so means only a fraction of the large fixed sized array is allocated.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3790", "sentence1": "A naively tiled code could perform that 8 times, requiring 64 total gets.", "sentence2": "an optimized tiling scheme for the same 8 steps would only require 27 get operations.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3791", "sentence1": "To address these problems, we developed an automated compiler support based on the latest advances in polyhedral frameworks (e.g., [5, 36]) to greatly reduce the human design effort currently required to create effectively synthesizable specification of designs using HLS tools.", "sentence2": "we developed compiler support for source-to-source transformations to optimize critical resources such as memory bandwidth to off-chip memory and on-chip buffer capacity [27].", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3792", "sentence1": "Standard tiling does not impose any constraint on tile sizes along spatial dimensions as a function of the time tile size.", "sentence2": "standard tiles may be compacted to a much smaller size to compensate for the larger tile sizes required by split-tiled dimensions.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3793", "sentence1": "Further, even once a defect has been detected and isolated, the creation of a small-scale reproducer can require precise understanding of the nature of the problem to preserve the salient characteristics.", "sentence2": "problems must often be largely resolved at-scale before a proper small-scale reproducer can be crafted.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3794", "sentence1": "The jump at 90% in the Y axis is due to one TCP timeout across both request and reply which is hard to avoid and is budgeted for (Figure 2).", "sentence2": "the network and compute budgets are in line with [5, 42, 44] and [39], respectively.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3795", "sentence1": "As we can see, the times spent on calculating the throughput of an application under ISPS and SPS are similar and much shorter than the time needed for solving the ILP problem to find the application throughput under PS and the time spent on finding the maximum achievable throughput of the application, i.e., the throughput under STS.", "sentence2": "our approach outperforms PS and STS in terms of time required to calculate the throughput of an application.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3796", "sentence1": "We assume that the ownship has the capability to perfectly estimate the geometrical configuration of the encounter, i.e., the relative position of the intruder, as well as the speed and direction of both aircraft; in practice such information is provided by different sensors as well as communication data exchanged be­tween aircraft.", "sentence2": "we discard all sources of noise and uncertainty related to the state estimation.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3797", "sentence1": "An important avenue of future work is to incorporate into the analysis the effects of numerical errors resulting from floating point calculations in the control code.", "sentence2": "aCSL utilizes the keywords requires and ensures to specify the preconditions and postconditions; the verification goal is to prove that postconditions are satisfied upon return if preconditions were satisfied when the function call occurred.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3798", "sentence1": "Unlike HDD, as flash memory stores and retrieves data by controlling electron movements, it has no seeking operations.", "sentence2": "there are fundamental questions that may be raised; that is \"what constrains the performance of SSD?\"", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3799", "sentence1": "If we copy a large file to a distributed file system, it will take a long time to copy the data to the target cluster compared to updating the directory.", "sentence2": "if many nodes copy large files to the same directory, then the amount of delay due to the concurrent directory update becomes relatively small compared to the amount of time to copy the file’s data.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3800", "sentence1": "In this case, TryBlocks can be retried locally and failures are detected and corrected lo­cally without any group-wise synchronization.", "sentence2": "a majority of data parallel applications need synchronization of some sort and any fault-tolerant API needs certain levels of synchronization.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3801", "sentence1": "To get an optimized I/O throughput on two-level storage, we use 1 MB request size for MapReduce applications and 4 MB I/O buffer between Tachyon and OrangeFS.", "sentence2": "this function is similar to the memory mountain that characterizes the capabilities of memory system.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3802", "sentence1": "The two­level storage can provide higher read and write throughput with limited number of compute nodes.", "sentence2": "running Hadoop with the two-level storage may provide a better performance solution for big data analytics on traditional HPC infrastructures.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3803", "sentence1": "Each of the files is setup to 1 GB.", "sentence2": "we use a single 10 GB file on RAMdisk in order to saturate the high bandwidth on the memory device.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3804", "sentence1": "Nonetheless, the use of C++ data structures did present problems.", "sentence2": "in order to avoid errors for both the PGI and Cray compilers, class data members used within parallel regions needed to be copied-in to the local scope.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3805", "sentence1": "As a result, some OpenACC compilers will succeed to automatically infer the range of array elements accessed within the parallel loops.", "sentence2": "for such compilers the data pragmas could be omitted to simplify the OpenACC parallel implementation.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3806", "sentence1": "When the MACC compiler encounters this clause, it generates reduction code at the same time as child kernel generation.", "sentence2": "the reduced variables must be returned to the parent kernel, using global memory to achieve this communication.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3807", "sentence1": "As illustrated in section 2.1, a MONC timestep is made up of a number of functionality groups and each group must run sequentially, i.e. the results from one group must be known before proceeding to the next group.", "sentence2": "it would not be efficient for us to port the entirety of the MONC dynamical core onto the GPU as the CPU would sit idle whilst the GPU is executing this aspect of the model.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3808", "sentence1": "These processes are not available until the job is finished.", "sentence2": "the processes in other worker programs and master program can con­tinue.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3809", "sentence1": "As of LLVM 3.6, LLVM IR supports a distinct property to prevent this merging [22], but OpenARC's LLVM backend currently targets LLVM 3.2.", "sentence2": "in other cases, the fault injection could cause the count to change, or it could cause the application to crash before reporting the count.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3810", "sentence1": "A FITL desert never has an entry portal and or an entry intrinsic.", "sentence2": "as in figure 4, the entry portals for FITL domains and FITL regions are always FITL deserts.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3811", "sentence1": "First, like our source-level fault injector, our FITL-based fault injector improves usability by supporting source-level pragmas.", "sentence2": "our FITL-based fault injector makes it easier to configure resilience studies and understand their results than when using lower-level fault injectors directly.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3812", "sentence1": "Space limita­tions prevent us from including more details on the locality analysis and optimization.", "sentence2": "for example, each array and variable can be either remote or local, and remote data operations can be expressed as just an assignment statement (e.g data = remoteData; in Figure 1).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3813", "sentence1": "In order to enable existing LLVM optimizations to apply to communication in PGAS programs, a PGAS programming language front-end needs to generate LLVM IR with the address space feature.", "sentence2": "the PGAS LLVM IR needs to include global pointers, which are an alternative way of representing a pointer to possibly remote memory and are recognizable by LLVM as pointer types, instead of the typical structure representation of wide pointers that include at least a node number and a local address.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3814", "sentence1": "The VFCache lookup could also have been merged with the call operation.", "sentence2": "virtual function targets and their references are relatively few in number, frequently reused, and rarely updated.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3815", "sentence1": "To fully use the VFCache, the processor must be able to identify which memory loads are, in fact, loading from a virtual function table suitable for caching with this new hardware structure.", "sentence2": "traditional ISAs make no special accommodations for virtual function tables.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3816", "sentence1": "A typical log message consists of timestamp, hostname, priority, facility, and an unstructured text message.", "sentence2": "more information may be extracted from the text.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3817", "sentence1": "In Excel, the Date/Time format specifies date and time to high precision, but the specification lacks information about time zones, thus it is not possible to convert time stamps between zones (e.g., local to UTC), nor is it possible to handle DST offsets. ", "sentence2": "using Excel as the storage format in large research projects requires maintenance of metadata for every data set containing time stamp values.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3818", "sentence1": "The data management module organizes necessary geospatial data for trigger modeling.", "sentence2": "each project corresponds to a project configuration file, where the information about the raster data such as topographic and fuel model data is stored.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3819", "sentence1": "The data management module organizes necessary geospatial data for trigger modeling.", "sentence2": "the 1 h, 10 h, and 100 h dead fuel moisture were set to 6%, and the live fuel moisture for wood and herbaceous fuel types were both set to 65%.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3820", "sentence1": "Moreover, as the size of the corpus increases, the number of nonzero elements in the co-occurrence matrix does not necessarily increase at the same pace.", "sentence2": "in some scenarios the product of the variety of different words and possible contexts would still exceed memory of a single machine.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3821", "sentence1": "Unfortunately, Go toolchain uses lightweight threads with minimal stacks that can grow dynamically with support of the Go runtime and external C libraries generally can't work on Go stack.", "sentence2": "a cgo function call requires switching of thread's stacks, which adds substantial performance overhead.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3822", "sentence1": "Any software that is not compiled with the AVX option will NOT use this instruction set and will be limited to performing SSE-type floating point instructions (and hence restricted to two flops per cycle).", "sentence2": "all are built with icc’s O3 level of optimization, AVX support, and icc’s ipo (inter-procedural optimization within and across source files).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3823", "sentence1": "Existing resource discovery approaches for IoT usu­ally rely upon a single repository responsible for storing resource descriptions [3].", "sentence2": "a centralized mechanism reveals several well-known drawbacks.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3824", "sentence1": "So­cial software allows quick access to information and easy means of networking.", "sentence2": "its laden with barriers such as redundancy with regard to other systems, a loss of control over one s data and knowledge, and its also somewhat time-consuming.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3825", "sentence1": "As pointed out by the focus group, enabling tools and structures for citizen involvement in service design processes must be developed, be they instruments allowing programming or piecing together concrete e-services, or ways for citizens to communicate their e-service proposals to the Public Sector Bodies or third-party developers, helping them to realize and implement their drafts.", "sentence2": "as we see it, the challenge does not lie that much in making citizens willing to participate in the extended CSLC, but to offer them the possibilities to do so.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3826", "sentence1": "In related work, concepts for context-aware workflows have been developed [3, 28] and proved to be applicable in practice.", "sentence2": "they suffer from the drawback that modelling context-aware workflows needs exponentially many activities and branches to handle all the possible context values and combinations.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3827", "sentence1": "The aim of GoI project is to mathematically understand dynamical process of cut-elimination.", "sentence2": "via Curry-Howard correspondence, GoI project aims to mathematically understand dynamical process of program execution.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3828", "sentence1": "However, in our case, the interpretation must be at a higher level: in the context of the application domain.", "sentence2": "we must explain what the program does in terms of the domain concepts.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3829", "sentence1": "In order to overcome the technological barrier, we have proposed an interaction mechanism based on physical objects (combining mobile devices and NFC Technology) and sensors (Bluetooth LE technology) to adapt the task (depending on the user's location) and provide non-intrusive interaction.", "sentence2": "the task adapts to user depending on the room.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3830", "sentence1": "Smart-homes are residences equipped with innovative technologies: sensors, wired and wireless networks, actuators and intelligent systems.", "sentence2": "in many cases, their human-machine interfaces are very difficult to use, especially for people with limitations, since voice-based devices, gesture commands in vision-based interfaces are artificial, unfamiliar, and easily forgotten.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3831", "sentence1": "This includes functions such as: environment control, care in the home and communication with the world around them.", "sentence2": "the system is capable of gathering: environmental parameters (temperature, luminosity, etc.) and security parameters (opening/closing of windows and doors, communication to/from family, friends, or medical staff, and even commands to devices such as on/off, up/down, etc.).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3832", "sentence1": "The OF-MMT home SDN domain controller is not able to compute such an inter-domain path, since it does not have the loca­tion information outside of its own controlled domain.", "sentence2": "the SDN domain controller needs to communicate, through east/westbound interface, with the neighbouring SDN do­main controllers to compute and setup network path towards CN.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3833", "sentence1": "Performance improves without cutoff on all runtime systems (Figure 1) since that provides suf.cient parallelism for our machine (Figure 11a).", "sentence2": "we generalize load balance to include tasks and display work deviation per-grain.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3834", "sentence1": "In this paper, we present language extensions for distributing pipelines and a new compiler backend that generates distributed code.", "sentence2": "the contributions of this paper are: A Halide language extension for distributing image processing pipelines, requiring programmers to write only approximately 10 new lines of code.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3835", "sentence1": "Consequently, there is more to UX than ever as the skills related to UX work have differentiated, new tools became available, and new ways of working, such as data-driven design, suggested.", "sentence2": "in addition to these elementary digital design skills, future would probably show demand for skills in data analytics and evidence-based design.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3836", "sentence1": "In previous work, the operating system or specialized hardware takes sole responsibility for power and energy efficiency.", "sentence2": "nRG-Loops enable applications to manage their own power and energy, resulting in more control for programmers and potentially more aggressive tradeoffs and therefore greater effciency.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3837", "sentence1": "In addition, a number of supplementary patterns were identified that should be documented in the future.", "sentence2": "developers may not trust your motivation as a company.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3838", "sentence1": "It focuses on the need to facilitate users to find the knowledge they need by better linking it to business processes and the related context.", "sentence2": "the reuse of knowledge can be facilitated, as well as its flows in the organisation.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3839", "sentence1": "Developers can use meta data (like annotations) to change some behavior of Scaffolding code generators, see lines 3 and 4 of Listing 5.", "sentence2": "in Scenario 1, the necessary widgets were completely unpredictable.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3840", "sentence1": "The automatically generated diverse replicas use the hardware differently, such that the same hardware fault leads to different consequences during their execution.", "sentence2": "the fault can be detected.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3841", "sentence1": "The timing is always influenced.", "sentence2": "how much the diversity technique affects the performance depends on the used method.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3842", "sentence1": "Hafiz et al. inspected the architecture of mail transfer agents.", "sentence2": "they compared the architecture of qmail in terms of an efficient security pattern usage in contrast to send mail [Hafiz et al. 2004].", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3843", "sentence1": "This search produced much output.", "sentence2": "we only considered the first 100 results presented at each search page due to increasing number of unfitting publications after 100 results.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3844", "sentence1": "Applications are accessed from a Browser, which has a Cache and interacts with the application through a Proxy Server.", "sentence2": "threat patterns are useful for developers because once they determine that a possible attack can happen in the environment, the pattern will indicate what security mechanisms are needed as countermeasures.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3845", "sentence1": "They may include misconfiguration exploits.", "sentence2": "this section describes the solution of the attacker s problem, i.e., how the attack can reach its objectives and the expected results of the attack.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3846", "sentence1": "After the modules are executed, the integrity violation can be located in any subsystem and may be hard to find.", "sentence2": "the execution of the malicious module should be prevented in the first place.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3847", "sentence1": "One of the challenges in large systems design is to recognize that it is the same entity or quantity (people usually know it, but tools don t), and to be able to reconcile the different characteristics and behaviors that may be assigned to it from different viewpoints.", "sentence2": "we are also able to create similar guidelines for modelling that enable at least manual reconciliation of the contents of models in different subdomains.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3848", "sentence1": "In addition, the use of Time Series to evaluate the refactoring solutions is another reason of the higher execution time.", "sentence2": "we can consider that an average of less than 25 minutes of difference between the execution time of our algorithm and existing work is not an issue especially that refactoring is not a real-time problem.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3849", "sentence1": "If the model well fits the data, the residuals of the model behave as an independent identically distributed sequence with a mean of zero and a variance of one.", "sentence2": "the residual sequence should correspond to a white noise; otherwise, the model needs to be improved.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3850", "sentence1": "The processing instances are build on top of the Amazon Elastic Compute Cloud web service that provides scalable on-demand usage-based pay as you go compute capacity.", "sentence2": "a platform for model storage using de facto standards.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3851", "sentence1": "For improving I/O performance in the virtualization environment, the minimum function of I/O stack, like merging, is necessary as the occasion demands.", "sentence2": "using virtio to eliminate I/O scheduler of the guest also has negative impacts on I/O performance.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3852", "sentence1": "It only stacks I/O requests in the virtqueue in the order of arrival.", "sentence2": "sorting is an unnecessary process and merely overhead for sequential write.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3853", "sentence1": "In work [6], a method for increasing I/O performance in virtualized environment by controlling block address in inode table in filesystem is proposed.", "sentence2": "the work aimed only to improve random access performance and does not mention sequential I/O throughput, HDD zones, and Hadoop performance.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3854", "sentence1": "Even before finishing, their load was not high.", "sentence2": "i/O utilization is not high at most times and CPU utilization is less than I/O utilization in Figure 4.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3855", "sentence1": "Of the 24 ratings of LondonTube, 21 exceeded the corresponding average rating of traditional tools; 2 ratings tied the corresponding average rating for traditional tools, and only 1 was lower.", "sentence2": "londonTube ratings exceeding traditional ratings outnumbered LondonTube ratings beneath traditional ratings by a ratio of 21-to-1 (Table 3).", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3856", "sentence1": "Patients can self-manage their drug use behavior by participating in self-help support groups [15, 18], such as Narcotics Anonymous (NA) [7], or through contact with experts and individuals with similar experience, such as the online forums on Reddit [9].", "sentence2": "many patients might still suffer from mild withdrawal symptoms in the early phase of this treatment stage and probably experience periods of relapse.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3857", "sentence1": "C Viewport Stabilization: we can make sure that when switching between versions a node will always have the exact same position in the viewport, represented here by the boxes around the nodes.", "sentence2": "the exceptional high standard error bars for the timeline condition in Task 3 and 6 can be explained by the different search strategies.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3858", "sentence1": "Usually, the same grouping mechanism is used for namespaces, classes, methods, loops, etc.", "sentence2": "we can approximate a rough syntax tree by only parsing code based on this grouping mechanism.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3859", "sentence1": "On the other hand, gaining performance at the expense of high energy consumption might not be practical for the system under design.", "sentence2": "metrics combining energy and performance have been proposed.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3860", "sentence1": "The EASE-DMF program takes a stone soup approach to funding in that each participant gains capabilities by providing a combination of funding and their own software capabilities to the soup.", "sentence2": "we would like to thank the Army Modeling and Simulation Office, the Army Research Lab Simulation and Training Technology Center, and the US Army Corps of Engineers Institute for Systems Engineering Research for their support of this program.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3861", "sentence1": "Because of the path explosion problem, full path-sensitive detection may cost greatly in analysis.", "sentence2": "partial path-sensitive detection are often applied in practical.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3862", "sentence1": "Our approach focuses on extracting user-specified web controls which act and appear the same as the one in the original context.", "sentence2": "web controls that are designed for desktop devices may not fit mobile screens of vary sizes well.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3863", "sentence1": "As records generated by different people may refer to the same entity, we need to deduplicate the record in entity resolution -Attribute: An entity is described by a set of attributes.", "sentence2": "further, we need to calculate the entropy of the entity's attribute.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3864", "sentence1": "Several of these companies have made substantial investments in the area of telehealth and many of them find that the public customers should just buy their solutions and get on with the job of providing telehealth services3 .", "sentence2": "this is not really happening.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3865", "sentence1": "We used insights obtained from our first analysis to se­lect the next incidents for analysis.", "sentence2": "after the first incident analysis we became cognisant of the different modalities in which data assets are prevalent in organisa­tions, namely (see Fig.  6(a)): data at rest (i.e., ‘inactive’ data stored in databases, file servers, archives, etc. that is not accessed or changed frequently), data in use (i.e., ‘active’ data processed by applications and held in computer memory, CPU caches/registers, or operational tables in databases) and data in motion (i.e., data in transit in the form of packets transported by communication protocols and data traversing the network, temporarily residing in memory buffers to be forwarded).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3866", "sentence1": "The ideal outcome for a mutation reduction strategy is to find the minimum set of mutants that can represent the complete set of mutants.", "sentence2": "our results suggest that the variance of individual samples is rather low, and the situation improves quite a bit with larger projects  — e.g. the variance of commons-math1 is just 0.397%.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3867", "sentence1": " Of course, one can still point out that random sampling is subject to the vagaries of chance, as one can get arbitrarily good or bad samples. ", "sentence2": "our results suggest that the variance of individual samples is rather low, and the situation improves quite a bit with larger projects — e.g. the variance of commons-math1 is just 0.397%.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3868", "sentence1": "In comparison with previous work [53, 54] our analysis is backed by theory and compares random sampling to the limit of selection.", "sentence2": "the results from our study are applicable to techniques such as clustering using static analysis, and even improved strata sampling techniques.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3869", "sentence1": "Portability requirements demand the system to be represented in a way that it .ts a specified environment, as for example the requirement The system shall run on platform X .", "sentence2": "furthermore, they found that at the pro ject level, NFRs are not taken into consideration during product planning (and are thereby not included as hard requirements in the projects) and they conclude that the realization of NFRs is a reactive rather than proactive effort.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3870", "sentence1": "Our results suggest that most non-functional requirements are not non-functional as they describe behavior of a system.", "sentence2": "we argue that many so-called NFRs can be handled similarly to functional requirements.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3871", "sentence1": "Furthermore, such real issue reports cannot be labeled by non-domain experts due to the lack of domain knowledge.", "sentence2": "it is very difficult for us to obtain a large-scale real dataset with high-quality labels.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3872", "sentence1": "An issue report also includes a time stamp recording the time at which the report was received.", "sentence2": "the issue report data can be treated as multi-dimensional, time series data.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3873", "sentence1": "Existing work in the area of energy-aware programming (e.g., [42, 60]) matches practitioners' focus on idle time by allowing computation to be degraded or delayed to save energy.", "sentence2": "to use such features effectively, practitioners must understand the energy impacts of their code, which given their lack of intuition (see Section 3.4), may be infeasible.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3874", "sentence1": "From a distributed systems theoretic standpoint, mitigating the results of asynchrony is impossible [42].", "sentence2": "integrators and contributors should agree on minimal communication protocols that increase each other's awareness and rendezvous points for mandatory information exchange.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3875", "sentence1": "we share the experimental data with the participants using an online survey platform, which forces the participants (1) to perform tasks in the desired order and (2) to fill in the questionnaires. ", "sentence2": "participants only got access to the final questionnaire after they had handed in their tasks, as well as they could not perform the second task without finishing the first one.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3876", "sentence1": "Having said this, ICFP papers analyzing human factors generally do not follow scienti.c conventions used in empirical dis­ciplines.", "sentence2": "null hypothesis H0-5: Experience level, defined by the position an individual has within an academic pipeline or professional sta­tus, has no impact on developer performance under any condition.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3877", "sentence1": "Further, given that our replication packet is trivial to port to other languages, our study simplifies the process of conducting such experiments for the community by providing a first baseline that can be adapted and tested under new conditions.", "sentence2": " the reader should keep in mind that no single experiment on non-trivial ideas can tell us the whole truth, which is one reason why the broader programming language wars [49] have been so difficult to evaluate.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3878", "sentence1": "The code samples did not show any way to do this other than the [&] so that participants would not be confused and delayed.", "sentence2": "our code samples can be considered generous for the lambda group, as we asked participants to use lambdas in only simple ways.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3879", "sentence1": "These handlers are indirectly called through event propagation mechanisms of JavaScript, where a single event can trigger multiple handlers of the ancestors of the target element [43].", "sentence2": "we capture propagated handlers and their relations with the original events.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3880", "sentence1": "Second, lists are more widely used than other collections such as maps or sets (Gson: 60%, K-9 Mail: 57%, Apache Com­mons Math: 56%, XStream: 50%, Apache Commons Con­.guration: 53%, Stock Exchange Trading Simulator: 57%).", "sentence2": "the energy contribution from lists is probably higher than that from other collections.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3881", "sentence1": "The dedicated site page stated that the issue (breach) had been resolved, Yes.", "sentence2": "to copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3882", "sentence1": "In a nutshell, the Tuckman model says that teams fight several conflicts before they start performing.", "sentence2": "team performance curves should show a certain behavior after the initial team setup, team reconfiguration, and over time.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3883", "sentence1": "To evaluate to which degree the implementation fulfills the requirements defined in Sec. 1, we performed a survey with\nthe students and measurements of the system’s performance.", "sentence2": "not all questions share the same number of answers.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3884", "sentence1": "One student only gave positive feedback, but graded the system with poor.", "sentence2": "we cannot rule out that students misunderstood the way they should score the system, leading to wrong marks.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3885", "sentence1": "These enable non-expert­programmers to build and maintain polished expert-like so­lutions, leveraging their domain expertise to express and compute with complex models.", "sentence2": "this flexibility presents a challenge for usage validation.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3886", "sentence1": "Indeed, the customer has to agree on what software they are paying for.", "sentence2": "it is also well recognized that often the customer is not clear on their real requirements [35]; it is the responsibility of the requirements engineer to help identify, clarify, and agree upon the actual requirements with the customer.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3887", "sentence1": "Ethics describes a gener­ally accepted set of moral principles, and addresses any intentional action that impacts negatively or positively the lives and values of others [21].", "sentence2": "ethics pro­vides moral guidance through principles; morals describe the goodness or badness of actions; values describe what an in­dividual or a group thinks is valuable or important [18, 1].", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3888", "sentence1": "By doing so, we are likely to incur a lock-in effect as we get tied in third-party software value­chains .", "sentence2": "there was a desire of mapping progress rather than anxiety incidents, for exam­ple by highlighting lack of interactions as a sign of progress in anxiety reduction.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3889", "sentence1": "When there is a need to update the local data through multiple asynchronous requests and it is required that all them succeed, an error on a single call can lead to having incorrect information shown to the user.", "sentence2": "developers need to explicitly implement proper recovery mechanisms, a task that most of times is complex and highly error prone, leading to tangled code and harder maintenance, especially in an asynchronous environment.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3890", "sentence1": "In code-centric development, one has to rely on manual code inspection for enforcing such policies - an error prone and time- and effort-intensive endeavor.", "sentence2": "in model-driven approach, such policies can be enforced either at model validation time or at model creation time itself.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3891", "sentence1": "the architectural layers are wired together by middleware infrastructure that supports message passing in a variety of protocols such as synchronous, asynchronous, publishsubscribe etc. ", "sentence2": "developing a distributed application demands wide-ranging expertise in distributed computing architectures and technology platforms.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3892", "sentence1": "A change in design decision, say, switching over to update-intensive behaviour needs only the database layer model to be transformed using a different model-to-text transformer.", "sentence2": "correct and consistent implementation of this change would be a huge challenge for code-centric approach.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3893", "sentence1": "The total time spent modeling ranges from ap­proximately 9 to 30 man hours, not counting the time spent by the researchers conducting the workshops.", "sentence2": "much of the time was spent on the type of discussions that many of the participants considered the most prominent benefit of the workshops.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3894", "sentence1": "As a system prototype, we have implemented a design tool which enables users to gather up every day objects and reassemble them to another functional shape with taking advantages of both analog and digital fabrication.", "sentence2": "the system calculates the optimized positional relationship among objects, and generates joint objects to bond the objects together in order to achieve a certain shape.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3895", "sentence1": "The researchers focused on situations where both parties interacted.", "sentence2": "they aimed to observe explicit or implicit negotiations of how security should be treated in software development and of who was to be responsible for taking on security-related tasks given their particular role in the team.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3896", "sentence1": "In consequence, the observations we make in experimental settings are only valid within those settings.", "sentence2": "the validity of the theories we build is strongly dependent on the contexts which are difficult to define because of the complex human, economical, technological, and cultural factors involved [14].", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3897", "sentence1": "To address this problem, a best practice-programming guide has been published by the OWASP foundation [9] to inform developers to match encoders to web contexts.", "sentence2": "there is no systematic way to test whether encoder functions are used correctly.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3898", "sentence1": "According to Moore [16], organizations often prefer to learn through the experiences of other organizations that belong to the same industry.", "sentence2": "organizations that are considering adopting DevOps can also benefit from a study that identifies the names of organizations that have adopted DevOps and are using software practices to integrate security.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3899", "sentence1": "The analysis of Internet artifacts and the survey indicated that, software practitioners find automated deployment to be beneficial to system's security.", "sentence2": "both the analysis of artifacts and survey revealed a security concern related to automated deployment if the DevOps organization uses improper automated deployment tools, or overlooks the need of security practices in the desire to deploy software rapidly.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3900", "sentence1": "We used the setup shown in Figure 7 to measure the energy consumption of an app on the mobile device.", "sentence2": "we used a Monsoon Power Monitor [37] to supply the mobile device with a steady voltage of 4.2 Volts and to measure its power consumption.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3901", "sentence1": "However, due to space restrictions, it would be impractical to discuss the design (& its re-factorings) for all of these 214 commits.", "sentence2": "we choose only 6 important commits (referred to as commits a to f), by observing a plot of changes (both GUI, as well as, source code), as shown in Figure 10.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3902", "sentence1": "This project was active for a period of approximately two years, in which duration it saw 214 commits.", "sentence2": "due to space restrictions, it would be impractical to discuss the design (& its re-factorings) for all of these 214 commits.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3903", "sentence1": "Essentially, any resource acquired during the execution of an app must be released before that app ceases to execute.", "sentence2": "since real-life apps have many potential exit locations, ensuring resource releases at all such exit locations may be challenging (specially for apps with many GUI screens).", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3904", "sentence1": "Another threat to validity to this work may arise due to the choice of subject programs.", "sentence2": "dynodroid uses a publicly-available, Android tool Hierarchy Viewer [11], to obtain the UI layout of an app.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3905", "sentence1": "The performance optimization of these complex stencils on current many-core architectures is full of significance.", "sentence2": "it is difficult and challenging to improve the performance of complex stencils on GPUs.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3906", "sentence1": "The original routing algorithm simply returns any shortest path between the two cores.", "sentence2": "abstract cost functions may lead to poor or infeasi­ble solutions.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3907", "sentence1": "Our approach to generating candi­dates is to expose this inherent tradeoff, using candidates of many different sizes then choosing the smallest canary from among the candidates that is similar enough to the full input according to one of the metrics described in §4.1.4.", "sentence2": "we assume only that the application devel­oper provides a well-defined accuracy calculation function.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3908", "sentence1": "The GC assumes the heap is contiguous, so the original blocks of these instances cannot remain empty.", "sentence2": "since a block's size is calculated from the class of the object within it, a block cannot remain an instance of the class as the class is now bigger than the block containing it, and traversing such a block would break the heap structure.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3909", "sentence1": "Together with test engineers from our industrial partner, we defined one cost measure, i.e., Time Difference (TD), which represents difference in period between a time budget and execution cost for a solution, and three effectiveness measures: Mean Priority (MPR), Mean Probability (MPO) and Mean Consequence (MC).", "sentence2": "in our context, each test case has four attributes: time, priority, probability, and consequence.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3910", "sentence1": "It may be tempting to think that it could have some efficiency benefit to let different writes (of different access patterns) operate on different views of an object.", "sentence2": "for GPU, there are no such needs in practice.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3911", "sentence1": "The model eliminates the possibility of dangling pointers and races, and ensures memory safety.", "sentence2": "this suggests that it is possible to have a fast GC implementation that also benefits from its implementation language's safety guarantees.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3912", "sentence1": "This is because GPUfs gread calls are blocking and use busy wait in GPU, wasting GPU resources due to the lack of an I/O preemption mechanism in GPUs.", "sentence2": "we employ opportunistic batching, which relies on the high rate of gread requests from the GPU, and naturally enables accumulating several requests while the previous batch is being transferred.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3913", "sentence1": "These algorithms use special locality-preserving hash functions, which map neighboring objects in high dimensional space into approximately neighboring objects in the low dimensional space.", "sentence2": "neighboring objects in the data set are likely to have the same hash key and therefore placed in the same bucket in a hash table.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3914", "sentence1": "Todays all-flash storage arrays exhibit excellent I/O throughput, latency, storage density, and energy efficiency.", "sentence2": "besides measuring the RBER, they also use the page program time and the PEC to build a linear model to predict the error rate of blocks.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3915", "sentence1": "Before writing data, the target location must be erased, an operation that can only be performed at block granularity.", "sentence2": "the data-placement unit draws freshly erased blocks from one of the Free Block Queues (FBQ) into the destage buffers, and separate destage buffers are maintained for each data stream.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3916", "sentence1": "In our prototype we implement coalesced threadblock-wide function calls, which are invoked by all the threads in a threadblock at the same point in the program, with the same arguments.", "sentence2": "multiple buffers are required to implement double buffering, and computations must be broken down to work on small tiles that fit into smaller buffers.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3917", "sentence1": "They must be mapped into GPU address space in order to be accessible from the GPU.", "sentence2": "the cudaHostRegister() call, which is used to map CPU physical memory into the GPU address space, does not support MMIO pages.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3918", "sentence1": "Kitten distinguished itself from their prior LWKs by providing a more complete Linux-compatible user environment.", "sentence2": "with the ever growing appetite for full Unix/POSIX feature compatibility from the application side, it has become increasingly difficult to support all these features without compromising the primary goal of LWK performance.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3919", "sentence1": "The maximum and minimum thresholds in Figure 14 are not expected to be exceeded.", "sentence2": "in the authors' proposition, two formal specifications (one written in algebraic notation and another using a synchronized transition system) describing the system are compared to each other, to help to understand and debug the informal requirements.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3920", "sentence1": "The approach is rather meant to validate certain system functionality with respect to the requirements.", "sentence2": "to check whether certain scenarios executed on the real system can be found in the state space of the formal model, which may indicate that the system implements the requirements as expected.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3921", "sentence1": "A special treatment is required to handle function calls in a device statement.", "sentence2": "a function must be compiled for the accelerator in order to make it accessible inside of a parallel region.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3922", "sentence1": "Scalar variables are implicitly private for an executing thread and need not be specified in a private clause in contrast to the presented OpenMP solution.", "sentence2": "the accessed matrices and their associated memory pointers must be passed in a present clause to omit a memory exchange at region entry.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3923", "sentence1": "2, where 20 tags are recorded with 40% sampling probability.", "sentence2": "one of them is to allow users to check the number of tags in an arbitrary tag flow passing through a distributed RFID system.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3924", "sentence1": "Quality drivers are the organizational motivations to invest in QM and when satisfied provide direct value to the organizational business objectives.", "sentence2": "5) Little guidance and assessment to automation whether process automation or automated execution.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3925", "sentence1": "Because we lack type class constraints in the language (and equality constraints over existentially-bound variables can easily be normalized away), we simply omit per­data-constructor constraints.", "sentence2": "typecase also performs run­time tests on type representations, and enables deconstructing type representations into their component parts for example, splitting a function type into a input type and output type.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3926", "sentence1": "Realizing such a contract relies on detecting calls to captured functions during the extent of a call to a closure.", "sentence2": "such proxies wrap carriers of contracts and intercept events as needed.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3927", "sentence1": "Finally, while the original work on functional ornaments used a reflective universe, we use type classes as an open-ended and extensible meta-programming framework.", "sentence2": "users are able to extend the framework at will, unlike the clients of a fixed reflective universe.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3928", "sentence1": "The third implementation is designed for the case of vertices with potentially many incoming edges.", "sentence2": "we focus on the dynamics of parallel computations in general, which includes async-finish and also other parallelism abstractions.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3929", "sentence1": "Sideways composition provides a technique to avoid some of the limitations of static, single inheritance object-oriented systems where multi-dimensional composition of behavior is desirable.", "sentence2": "context-oriented programming applies sideways composition to improve modularity.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3930", "sentence1": "The interaction layer is in charge of gathering information from the system's surrounding environment.", "sentence2": "information coming from the physical environment (e.g., geographical localisation), the user (e.g., user preferences), or the system's computing platform (i.e., hardware and software conditions, such as the battery status).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3931", "sentence1": "Functional Hazard Analysis is the approach that is close to the approach presented below [13].", "sentence2": "it does take a set of known functions as its starting point.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3932", "sentence1": "A proof certificate is similar to a proof object, but it can avoid recording all the needed information or encode it at a different level of detail.", "sentence2": " a proof certificate may be non syntax directed, in the sense that multiple rules can be tried according to the same certificate, triggering backtracking if needed.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3933", "sentence1": "We observe that the tactics and tacticals that can be implemented in our system go beyond the LCF ones because we inherit full-backtracking from λProlog.", "sentence2": "our tactics are naturally non-deterministic, i.e. they can have multiple solutions (that are enumerated via backtracking).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3934", "sentence1": "This difference makes our approach actually create new functions and insert function calls to those newly created functions in the source code.", "sentence2": "in the previous techniques for on-chip memory management [33, 14] the created partitions are still part of the existing func­tion; branches at partition boundaries need to be modified to maintain the original control flow even when the partitions are loaded to nonconsecutive memory ranges.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3935", "sentence1": "New tunneling protocols are being proposed on a daily basis.", "sentence2": "a method specialized in a particular protocol will soon be out of date.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3936", "sentence1": "Through playtesting it was determined that of the implemented methods, map stations were the most effective in giving an overview of the game state.", "sentence2": "in future iterations other methods of giving situation awareness will be tested.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3937", "sentence1": "We interpret this figure just like Figure 4.", "sentence2": "clearly, a smartphone may be in a variety of states, but for our purposes we are effectively treating a screen event as a state.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3938", "sentence1": "Next, our analysis considers the battery level of the phone when a state transition takes place.", "sentence2": "at the moment a transition was recorded by our software, we also noted the battery level at that moment.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3939", "sentence1": "World of Riders is intended to target casual players of both genders and various age groups who share attitudes of healthy lifestyle and regular exercising.", "sentence2": "given the scope of the present project, it was addressed the audience who generally support the idea that regular exercising has a positive effect on their health, but in reality experience hard times in maintaining their level of motivation and engagement in performing such physical activities on a regular basis.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3940", "sentence1": "We will not detail all changes here, we just mention what we had to customize: variable declaration syntax in Xbase is different from Java so we customized the corresponding grammar rule; more­over, Xbase does not support several variable declarations in a single statement as in Java.", "sentence2": "we must handle possible additional variable declarations in Jbase.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3941", "sentence1": "Finally, Xtext provides complete support for typical Java build tools, like Maven and Gradle.", "sentence2": "xtext DSLs also automatically support these build tools.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3942", "sentence1": "As we showed in the examples of Section 4, it is straightforward with Jbase to introduce Java expressions and statements into an Xtext DSL.", "sentence2": "we believe that Jbase is a valuable tool also for extending Java.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3943", "sentence1": "Users inspect the suggested annotation sets for soundness and can ask AUTOBAHN to automatically patch their program sources.", "sentence2": "the data also shows that AUTOBAHN annota­tions outperform the hand annotations for gcSimulator: the geometric mean of gcSimulator running times when optimized by AUTOBAHN is 58.6% of the bare runtime, compared to a geomet­ric mean of 64.8% of bare for the hand-optimized version.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3944", "sentence1": "They rely on a CPS-like interpretation of session­types in terms of one-shot (or linear) channels, which they can im­plement using Scala's Future type.", "sentence2": "unlike many presentations of session types, but inspired by their logical connections, our session types represent closing of channels explicitly.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3945", "sentence1": "One important aspect of a modern language runtime is warmup performance.", "sentence2": "how the performance of the language runtime evolves after initialization and during justin-time compilation.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3946", "sentence1": "It is remarkable that despite this, we can still observe speedups up to a factor 10.", "sentence2": "although non reproducible, we consider these measure­ments much more important in practice, since they are real life usages of the optimizer.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3947", "sentence1": "A fundamental challenge then is coming up with a probabilistic model of code which is both general and precise.", "sentence2": "building such model is a difficult problem and one which has recently attracted the attention of researchers from several areas.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3948", "sentence1": "GRANDET always incurs less than 0.2ms latency, smaller than the standard deviation of latency for each case.", "sentence2": "we modeled the usage data of Cumulus-Clips according to YouTube [51], where a billion users upload 300 hours of video per minute [40] and an average video has four minutes [14, 42].", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3949", "sentence1": "Thus, given that the number of rewrite operations for STT-MRAM is a perfectly feasible number for main memory or CPU cache, STT-MRAM has the potential of serving as the main memory of computer systems like DRAM does today.", "sentence2": "one disadvantage of STT-MRAM is that the write energy will be 102 times larger than that of DRAM.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3950", "sentence1": "We propose that the hypervisor layer, located between guest-OS software and physical hardware, is most suitable for implementing page placement optimization.", "sentence2": "existing hypervisors designed for DRAM-only memory architecture, do not have any support for hybrid memory systems; there is no mechanism to allocate guest pages from DRAM and NVM pages and dynamically optimize page mappings.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3951", "sentence1": "In particular, the partitions from one iteration to the next are quite similar, thus, our smart partitioning assignment has significant benefits.", "sentence2": "rUBiS reveals that as soon as the workload and data design becomes slightly more complex, the generated graphs and resulting partitions differ a lot from one iteration to the next.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3952", "sentence1": "Finally, in each iteration, we determine requests and objects that can be pruned.", "sentence2": "we ignore requests and objects that have fading popularity and we ignore requests that have too high drift rate to benefit from caching.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3953", "sentence1": "New policies are sent to the load-balancer and the object caches whenever needed.", "sentence2": "in order to not miss these situations, we trigger a new reconfiguration after a certain, fairly long time, even if the degradation threshold has not been reached.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3954", "sentence1": "Although a request type might access two objects, not every call will access these two objects.", "sentence2": "we use as edge weight the sum of both objects access frequencies caused by requests that access both objects.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3955", "sentence1": "In addition, they use several optimizations to enhance the quality of the produced partitions.", "sentence2": "they produce partitions with different content every time, and therefore, there is less overlap with the current object distribution.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3956", "sentence1": "This value is the total number of movements invoked by placing SSi at SLOTj .", "sentence2": "orion not only improves performance up to 1.61 times speedup but also resource &#38; energy efficiency, with up to 62.5% memory resource saving, and 6.7% energy reduction over the highly optimized code gen­erated by nvcc (Section 4).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3957", "sentence1": "For exam­ple, the snippet might wait on a signal from a different thread to continue the execution.", "sentence2": "[7] focuses on mining examples from unit tests claiming that this is a good source of examples as they are concise, relevant and trustworthy.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3958", "sentence1": "We thus argue that RPython is a more friendly VM development platform for novices than Slang/VMMaker.", "sentence2": "many optimizations and platform interfaces can be developed using high-level RPython application programming interfaces (APIs).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3959", "sentence1": "Building on existing work by speculative designers and artists, this project opens opportunities for further research in HCI.", "sentence2": "further research is needed to understand the role of interactive objects in helping the audience understand complex issues and create alternative narratives about the future.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3960", "sentence1": "The goal of this project is to demonstrate how speculative design and social media can be used in public discourse and suggest new avenues for research at the intersection of speculative design and HCI.", "sentence2": "it is far from the completely dystopian and authoritarian future George Orwell portrayed in 1984.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3961", "sentence1": "Dewey also defined an experience as a balance between doing and undergoing [9].", "sentence2": "experience is not something we receive passively but participate in the construction of it.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3962", "sentence1": "The overall goal of the workshop was to create a shared understanding of UX, provoke discussion and encourage rethinking and reflection, engage the employees to the design cultural change, co-create guidelines for empowerment, and ideate possible ways to implement them into the design process.", "sentence2": "a participatory workshop was conducted in order to prepare the employees to accept this still quite unfamiliar and sometimes confusing methodology of UX that was only slightly familiar to them.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3963", "sentence1": "Yet, age-associated cognitive changes are not irreversible and can be improved with adequate training [I].", "sentence2": "cognitive stimulation is a component of \"active aging\", a term was adopted by the World Health Organization (WHO) to promote a better quality of life and improved autonomy and independence of older people.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3964", "sentence1": "As a result, directly modifying a package declaration itself is currently not possible in PEoPL.", "sentence2": " we wanted to show that PEoPL can indeed handle all DeltaJ operations, and thus left the product line as is (cf. [33]).", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3965", "sentence1": "To ensure isolation of the different actors, objects that are transmitted over a mailbox have to implement the Message interface and are passed by copy.", "sentence2": "there exist extensions to Kilim's type system to enable zero-copy message passing [10].", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3966", "sentence1": "However, we found in experiences that also these people and their skill profiles are influenced by the overall ecosystem context.", "sentence2": "we summarize further roles in this section with their commonly desired skills.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3967", "sentence1": "The results of the proof-of-concept empirical study, i.e., which of the afore-mentioned mappings have been identified in the explored games, are outlined in Table I.", "sentence2": "the class diagram created for this mapping consists of an abstract Game class whose member function gameloop plays the role of the template method and defines the structure of the al-gorithm that implements the game loop.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3968", "sentence1": "In this experiment, we plan to implement an BCI (Brain Computer Interaction) interface, where the user will be able to change the from an avatar to another by measuring his/her brain activity.", "sentence2": "if the User relaxes, her/his brain activity is in low level and its avatar morphs to a more gentle or friendly avatar, if the User is in alert, his/her brain activity is in high level and its avatar morphs to a more intimidating or hostile avatar.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3969", "sentence1": "For example, all the building should contain the fireproofing rating property.", "sentence2": "and it is difficult for the domain experts who are not proficient in computer programming to write code for automatic checking.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3970", "sentence1": "In this paper, by using a visualization tool that we developed, the users may view this tool as helpful because of selection of many sentiment analysis reviews in the list, not reading all the reviews of the existing list.", "sentence2": "scores also obtained by 3 points from negative words such as \"evil\".", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3971", "sentence1": "System must extract one or more ASIN(Amazon Standard Identification Number) of the movie title by user's choice[10].", "sentence2": "system obtains ASIN from Amazon ASIN Database that is established in advance.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3972", "sentence1": "Each miner receiving a newly cre­ated block validates the block's integrity.", "sentence2": "it makes sure that (i) the block's SHA-2562 is indeed lower than the difficulty threshold, (ii) all transactions in it are legitimate (i.e., if all input coins are in the UTxO set and are signed by their claimed owners), and that (iii) the miner that created it has allocated the correct amount to his own wallet (i.e., the number of bitcoins offered per created block, plus the fees of all included transactions).", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3973", "sentence1": "A block is indexed in the DHT by its own hash value.", "sentence2": "we maintain an array of all blocks’ hashes, which we call the index.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3974", "sentence1": "Request graphs tend to be smaller as the number of different requests is considerably smaller than the number of different objects.", "sentence2": "every cache is aware of other cache contents using a replicated directory.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3975", "sentence1": "The underlying idea of business domain orientation, fostered by these patterns, is also an integral part of the Star pattern.", "sentence2": "in contrast to the Star pattern, where representational connectors are mandatory, no such building block exists in DDD.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3976", "sentence1": "Performance The service should be capable of delivering high-quality audio within acceptable latency bounds.", "sentence2": "the latency between audio device sources should be less than that of the human hearing detectability threshold of 50ms for speech [6].", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3977", "sentence1": "There is no evidence there is a difference between the underlying mean latencies for the different number of channels in surround sound mode (p-value=0.12).", "sentence2": "for the number of channels tested in this experiment, it cannot be claimed that the number of channels has an effect on listener to listener latency.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3978", "sentence1": "In contrast, a careful discussion is needed for isEmpty, as there is no computable counterpart of isEmpty in general functional programs, and it must be converted to an productivity-test oracle (Section 4).", "sentence2": "unlike that in Section 4, the discussions in Sections 5, 6 and 7 can be easily extended to isEmpty because isEmpty becomes computable for FUnCAL programs thanks to the type system in Section 6.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3979", "sentence1": "Those components, which are primarily developed as a part of applications, are integrated within the modeling tool.", "sentence2": "several end-user applications, generated in the run-time, are simultaneously directed towards single target RTS and instance of a program being debugged.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3980", "sentence1": "The TCP/IP channels, or remote consoles are used to access remote RTSs.", "sentence2": "in the model-level debugging, purpose of the sub-model construction is to: 1) reduce the set of application states to the level that it can be visually traced; 2) vary values of object included in a model in an efficient and simple way; and 3) generate representative documentation.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3981", "sentence1": "The visual graph-node-arc convention surreptitiously introduces the user to the concept of type by requiring him to match the type of the output of an upstream node with the type expected as input to a downstream node.", "sentence2": "visual programming can become extremely verbose and does not scale to more complex programing tasks.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3982", "sentence1": "Vaniea et al. [40] already found that users rely on memories of previous update experiences when making future update decisions and that the duration of upgrades seems to have some impact.", "sentence2": "we wanted to better investigate the link between actual experience, memory, and future decisions.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3983", "sentence1": "Participants expressed an interest in hoarding code through commenting out code snippets and refusing to delete old source code files in case their exploration did not pay off.", "sentence2": "this lead to confusion with keeping track of multiple similar copies as well as commented versus not-commented code.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3984", "sentence1": "Unlike conventional version control, these informal practices allow for fast versioning of any size code snippet, and quick comparisons by interchanging which versions are run.", "sentence2": "data scientists must maintain a strong mental map of their code in order to distinguish versions, leading to errors and confusion.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3985", "sentence1": "While viewing an earlier commit, El­len cannot edit the code, but she can re-run it, copy it, or cre­ate a new branch from that point in time.", "sentence2": "data scientists must maintain a strong mental map of their code in order to distinguish versions, leading to errors and confusion.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3986", "sentence1": "In this paper, we present a new empirical study that both addresses our goal of principles as well as overcomes the lim­itations of prior Patchworks evaluations.", "sentence2": "we performed an observational user study that compared the ubiq­uitous tabbed editor design with the experimental Patchworks design to inductively build a set of design principles.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3987", "sentence1": "Non-temporary assignment is a separate in­struction, because it involves complex logic in the general case (de­stroying the previous value safely, handling reference assignments, handling overloaded assignment operators, etc.)", "sentence2": "no bytecode representation is used, instead all operations are performed on the AST level.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3988", "sentence1": "For example, the Starkiller project [20] translates Python code to C++, using an augmented Cartesian product algorithm [21] for type inference.", "sentence2": "this approach is often not able to support all language semantics.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3989", "sentence1": "[23] compiles a functional language (Scheme) into a dataflow representation to produce custom hardware.", "sentence2": "they only implement strict functions and do not exploit the pipelined parallelism we can.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3990", "sentence1": "Professional, novice, or amateur designers, are now equipped with the means to fabricate and test ideas rapidly at a relatively low cost, and at a much earlier stage of the design process.", "sentence2": "just like any new tools, integrating digital fabrication into the design process is often a trial-and-error process such as specifying the right joints and parameters for different materials during laser cutting.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3991", "sentence1": "In every field, the correct tools are needed to complete a job or tasks.", "sentence2": "in mobile digital forensics, software is one of the most valuable tools.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3992", "sentence1": "When the GPU driver receives a request for GPU memory that cannot be satisfied, GPrioSwap first selects a victim which must give up some of its memory.", "sentence2": "either the user space libraries or the applications themselves must be modified to support these new APIs.", "label": "neutral", "domain": "Software and its engineering"}
{"id": "test_3993", "sentence1": "To that end, we need to count accesses to individual pages with sufficient accuracy.", "sentence2": "as we explain in Section 2, current GPUs do not support counting accesses to individual pages out of the box.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3994", "sentence1": "This hybrid approach was informed by earlier findings on the design of block-based tools, including features that learners found to be useful as well as perceived drawbacks of block-based programming tools [36].", "sentence2": "this hybrid design retains features including a browsable blocks library, drag-and-drop composition, and pre-fabricated commands.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3995", "sentence1": "Secondly, maybe students contribute only for a short period and then drop out of the project, while employed contributors stick with the project after their first contributions.", "sentence2": "later samples of OSS contributors would cover the employed contributors who are still with the project, but not the students who had left after their first contributions already.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3996", "sentence1": "Core contribution barriers like Solution are obviously necessary for the modification and therefore less likely to be underestimated; thus, we operationalize the unexpected effort via the secondary contribution barriers Download, Setup, and Build. ", "sentence2": "there are no statistically significant differences for these contribution barriers.", "label": "contrasting", "domain": "Software and its engineering"}
{"id": "test_3997", "sentence1": "An additional wrinkle is that it is much more difficult to find spam projects on SourceForge using the downloads metric, as we did with CodePlex, due to inconsistencies in the way data was donated to SRDA over the years [13], and the lamentable fact that there seems to be no accurate mechanism for querying downloads after 2009.", "sentence2": "we differentiate between all projects and and likely nonspam projects using the amount of text in the description field, since legitimate projects will include a description, even if it is short.", "label": "reasoning", "domain": "Software and its engineering"}
{"id": "test_3998", "sentence1": "By looking at the results reported by the accuracy measure and F1­measure, we can see a discrepancy between them, possibly due to the sample size and, more importantly, our use of 10-way cross validation, which usually overestimates errors in models [20].", "sentence2": "the real performance of the models in F1-measure exceeds the values reported in Table 6.", "label": "entailment", "domain": "Software and its engineering"}
{"id": "test_3999", "sentence1": "This result indicates that algorithm factor contributes to increased pricing.", "sentence2": "it might reveal more complexity or some constraints within the task.", "label": "entailment", "domain": "Software and its engineering"}