source
stringlengths 200
2.98k
| target
stringlengths 18
668
|
|---|---|
Graph Convolutional Networks (GCNs) have recently been shown to be quite successful in modeling graph-structured data. However, the primary focus has been on handling simple undirected graphs. Multi-relational graphs are a more general and prevalent form of graphs where each edge has a label and direction associated with it. Most of the existing approaches to handle such graphs suffer from over-parameterization and are restricted to learning representations of nodes only. In this paper, we propose CompGCN, a novel Graph Convolutional framework which jointly embeds both nodes and relations in a relational graph. CompGCN leverages a variety of entity-relation composition operations from Knowledge Graph Embedding techniques and scales with the number of relations. It also generalizes several of the existing multi-relational GCN methods. We evaluate our proposed method on multiple tasks such as node classification, link prediction, and graph classification, and achieve demonstrably superior results. We make the source code of CompGCN available to foster reproducible research. | A Composition-based Graph Convolutional framework for multi-relational graphs. |
State-of-the-art neural machine translation methods employ massive amounts of parameters. Drastically reducing computational costs of such methods without affecting performance has been up to this point unsolved. In this work, we propose a quantization strategy tailored to the Transformer architecture. We evaluate our method on the WMT14 EN-FR and WMT14 EN-DE translation tasks and achieve state-of-the-art quantization results for the Transformer, obtaining no loss in BLEU scores compared to the non-quantized baseline. We further compress the Transformer by showing that, once the model is trained, a good portion of the nodes in the encoder can be removed without causing any loss in BLEU. | We fully quantize the Transformer to 8-bit and improve translation quality compared to the full precision model. |
Gradient-based meta-learning techniques are both widely applicable and proficient at solving challenging few-shot learning and fast adaptation problems. However, they have practical difficulties when operating on high-dimensional parameter spaces in extreme low-data regimes. We show that it is possible to bypass these limitations by learning a data-dependent latent generative representation of model parameters, and performing gradient-based meta-learning in this low-dimensional latent space. The resulting approach, latent embedding optimization (LEO), decouples the gradient-based adaptation procedure from the underlying high-dimensional space of model parameters. Our evaluation shows that LEO can achieve state-of-the-art performance on the competitive miniImageNet and tieredImageNet few-shot classification tasks. Further analysis indicates LEO is able to capture uncertainty in the data, and can perform adaptation more effectively by optimizing in latent space. | Latent Embedding Optimization (LEO) is a novel gradient-based meta-learner with state-of-the-art performance on the challenging 5-way 1-shot and 5-shot miniImageNet and tieredImageNet classification tasks. |
We introduce an approach for augmenting model-free deep reinforcement learning agents with a mechanism for relational reasoning over structured representations, which improves performance, learning efficiency, generalization, and interpretability. Our architecture encodes an image as a set of vectors, and applies an iterative message-passing procedure to discover and reason about relevant entities and relations in a scene. In six of seven StarCraft II Learning Environment mini-games, our agent achieved state-of-the-art performance, and surpassed human grandmaster-level on four. In a novel navigation and planning task, our agent's performance and learning efficiency far exceeded non-relational baselines, it was able to generalize to more complex scenes than it had experienced during training. Moreover, when we examined its learned internal representations, they reflected important structure about the problem and the agent's intentions. The main contribution of this work is to introduce techniques for representing and reasoning about states in model-free deep reinforcement learning agents via relational inductive biases. Our experiments show this approach can offer advantages in efficiency, generalization, and interpretability, and can scale up to meet some of the most challenging test environments in modern artificial intelligence. | Relational inductive biases improve out-of-distribution generalization capacities in model-free reinforcement learning agents |
Image translation between two domains is a class of problems aiming to learn mapping from an input image in the source domain to an output image in the target domain. It has been applied to numerous applications, such as data augmentation, domain adaptation, and unsupervised training. When paired training data is not accessible, image translation becomes an ill-posed problem. We constrain the problem with the assumption that the translated image needs to be perceptually similar to the original image and also appears to be drawn from the new domain, and propose a simple yet effective image translation model consisting of a single generator trained with a self-regularization term and an adversarial term. We further notice that existing image translation techniques are agnostic to the subjects of interest and often introduce unwanted changes or artifacts to the input. Thus we propose to add an attention module to predict an attention map to guide the image translation process. The module learns to attend to key parts of the image while keeping everything else unaltered, essentially avoiding undesired artifacts or changes. The predicted attention map also opens door to applications such as unsupervised segmentation and saliency detection. Extensive experiments and evaluations show that our model while being simpler, achieves significantly better performance than existing image translation methods. | We propose a simple generative model for unsupervised image translation and saliency detection. |
Building deep neural networks to control autonomous agents which have to interact in real-time with the physical world, such as robots or automotive vehicles, requires a seamless integration of time into a network’s architecture. The central question of this work is, how the temporal nature of reality should be reflected in the execution of a deep neural network and its components. Most artificial deep neural networks are partitioned into a directed graph of connected modules or layers and the layers themselves consist of elemental building blocks, such as single units. For most deep neural networks, all units of a layer are processed synchronously and in parallel, but layers themselves are processed in a sequential manner. In contrast, all elements of a biological neural network are processed in parallel. In this paper, we define a class of networks between these two extreme cases. These networks are executed in a streaming or synchronous layerwise-parallel manner, unlocking the layers of such networks for parallel processing. Compared to the standard layerwise-sequential deep networks, these new layerwise-parallel networks show a fundamentally different temporal behavior and flow of information, especially for networks with skip or recurrent connections. We argue that layerwise-parallel deep networks are better suited for future challenges of deep neural network design, such as large functional modularized and/or recurrent architectures as well as networks allocating different network capacities dependent on current stimulus and/or task complexity. We layout basic properties and discuss major challenges for layerwise-parallel networks. Additionally, we provide a toolbox to design, train, evaluate, and online-interact with layerwise-parallel networks. | We define a concept of layerwise model-parallel deep neural networks, for which layers operate in parallel, and provide a toolbox to design, train, evaluate, and on-line interact with these networks. |
Deep neural networks are known to be vulnerable to adversarial perturbations. In this paper, we bridge adversarial robustness of neural nets with Lyapunov stability of dynamical systems. From this viewpoint, training neural nets is equivalent to finding an optimal control of the discrete dynamical system, which allows one to utilize methods of successive approximations, an optimal control algorithm based on Pontryagin's maximum principle, to train neural nets. This decoupled training method allows us to add constraints to the optimization, which makes the deep model more robust. The constrained optimization problem can be formulated as a semi-definite programming problem and hence can be solved efficiently. Experiments show that our method effectively improves deep model's adversarial robustness. | An adversarial defense method bridging robustness of deep neural nets with Lyapunov stability |
In this paper, we propose a method named Dimensional reweighting Graph Convolutional Networks (DrGCNs), to tackle the problem of variance between dimensional information in the node representations of GCNs. We prove that DrGCNs can reduce the variance of the node representations by connecting our problem to the theory of the mean field. However, practically, we find that the degrees DrGCNs help vary severely on different datasets. We revisit the problem and develop a new measure K to quantify the effect. This measure guides when we should use dimensional reweighting in GCNs and how much it can help. Moreover, it offers insights to explain the improvement obtained by the proposed DrGCNs. The dimensional reweighting block is light-weighted and highly flexible to be built on most of the GCN variants. Carefully designed experiments, including several fixes on duplicates, information leaks, and wrong labels of the well-known node classification benchmark datasets, demonstrate the superior performances of DrGCNs over the existing state-of-the-art approaches. Significant improvements can also be observed on a large scale industrial dataset. | We propose a simple yet effective reweighting scheme for GCNs, theoretically supported by the mean field theory. |
Knowledge-grounded dialogue is a task of generating an informative response based on both discourse context and external knowledge. As we focus on better modeling the knowledge selection in the multi-turn knowledge-grounded dialogue, we propose a sequential latent variable model as the first approach to this matter. The model named sequential knowledge transformer (SKT) can keep track of the prior and posterior distribution over knowledge; as a result, it can not only reduce the ambiguity caused from the diversity in knowledge selection of conversation but also better leverage the response information for proper choice of knowledge. Our experimental results show that the proposed model improves the knowledge selection accuracy and subsequently the performance of utterance generation. We achieve the new state-of-the-art performance on Wizard of Wikipedia (Dinan et al., 2019) as one of the most large-scale and challenging benchmarks. We further validate the effectiveness of our model over existing conversation methods in another knowledge-based dialogue Holl-E dataset (Moghe et al., 2018). | Our approach is the first attempt to leverage a sequential latent variable model for knowledge selection in the multi-turn knowledge-grounded dialogue. It achieves the new state-of-the-art performance on Wizard of Wikipedia benchmark. |
Meta-learning, or learning-to-learn, has proven to be a successful strategy in attacking problems in supervised learning and reinforcement learning that involve small amounts of data. State-of-the-art solutions involve learning an initialization and/or learning algorithm using a set of training episodes so that the meta learner can generalize to an evaluation episode quickly. These methods perform well but often lack good quantification of uncertainty, which can be vital to real-world applications when data is lacking. We propose a meta-learning method which efficiently amortizes hierarchical variational inference across tasks, learning a prior distribution over neural network weights so that a few steps of Bayes by Backprop will produce a good task-specific approximate posterior. We show that our method produces good uncertainty estimates on contextual bandit and few-shot learning benchmarks. | We propose a meta-learning method which efficiently amortizes hierarchical variational inference across training episodes. |
Often we wish to transfer representational knowledge from one neural network to another. Examples include distilling a large network into a smaller one, transferring knowledge from one sensory modality to a second, or ensembling a collection of models into a single estimator. Knowledge distillation, the standard approach to these problems, minimizes the KL divergence between the probabilistic outputs of a teacher and student network. We demonstrate that this objective ignores important structural knowledge of the teacher network. This motivates an alternative objective by which we train a student to capture significantly more information in the teacher's representation of the data. We formulate this objective as contrastive learning. Experiments demonstrate that our resulting new objective outperforms knowledge distillation on a variety of knowledge transfer tasks, including single model compression, ensemble distillation, and cross-modal transfer. When combined with knowledge distillation, our method sets a state of the art in many transfer tasks, sometimes even outperforming the teacher network. | Representation/knowledge distillation by maximizing mutual information between teacher and student |
Developing effective biologically plausible learning rules for deep neural networks is important for advancing connections between deep learning and neuroscience. To date, local synaptic learning rules like those employed by the brain have failed to match the performance of backpropagation in deep networks. In this work, we employ meta-learning to discover networks that learn using feedback connections and local, biologically motivated learning rules. Importantly, the feedback connections are not tied to the feedforward weights, avoiding any biologically implausible weight transport. It can be shown mathematically that this approach has sufficient expressivity to approximate any online learning algorithm. Our experiments show that the meta-trained networks effectively use feedback connections to perform online credit assignment in multi-layer architectures. Moreover, we demonstrate empirically that this model outperforms a state-of-the-art gradient-based meta-learning algorithm for continual learning on regression and classification benchmarks. This approach represents a step toward biologically plausible learning mechanisms that can not only match gradient descent-based learning, but also overcome its limitations. | Networks that learn with feedback connections and local plasticity rules can be optimized for using meta learning. |
In the visual system, neurons respond to a patch of the input known as their classical receptive field (RF), and can be modulated by stimuli in the surround. These interactions are often mediated by lateral connections, giving rise to extra-classical RFs. We use supervised learning via backpropagation to learn feedforward connections, combined with an unsupervised learning rule to learn lateral connections between units within a convolutional neural network. These connections allow each unit to integrate information from its surround, generating extra-classical receptive fields for the units in our new proposed model (CNNEx). We demonstrate that these connections make the network more robust and achieve better performance on noisy versions of the MNIST and CIFAR-10 datasets. Although the image statistics of MNIST and CIFAR-10 differ greatly, the same unsupervised learning rule generalized to both datasets. Our framework can potentially be applied to networks trained on other tasks, with the learned lateral connections aiding the computations implemented by feedforward connections when the input is unreliable. | CNNs with biologically-inspired lateral connections learned in an unsupervised manner are more robust to noisy inputs. |
Deep learning (DL) has in recent years been widely used in natural
language processing (NLP) applications due to its superior
performance. However, while natural languages are rich in
grammatical structure, DL has not been able to explicitly
represent and enforce such structures. This paper proposes a new
architecture to bridge this gap by exploiting tensor product
representations (TPR), a structured neural-symbolic framework
developed in cognitive science over the past 20 years, with the
aim of integrating DL with explicit language structures and rules.
We call it the Tensor Product Generation Network
(TPGN), and apply it to image captioning. The key
ideas of TPGN are: 1) unsupervised learning of
role-unbinding vectors of words via a TPR-based deep neural
network, and 2) integration of TPR with typical DL architectures
including Long Short-Term Memory (LSTM) models. The novelty of our
approach lies in its ability to generate a sentence and extract
partial grammatical structure of the sentence by using
role-unbinding vectors, which are obtained in an unsupervised
manner. Experimental results demonstrate the effectiveness of the
proposed approach. | This paper is intended to develop a tensor product representation approach for deep-learning-based natural language processinig applications. |
It is well-known that classifiers are vulnerable to adversarial perturbations. To defend against adversarial perturbations, various certified robustness results have been derived. However, existing certified robustnesses are limited to top-1 predictions. In many real-world applications, top-$k$ predictions are more relevant. In this work, we aim to derive certified robustness for top-$k$ predictions. In particular, our certified robustness is based on randomized smoothing, which turns any classifier to a new classifier via adding noise to an input example. We adopt randomized smoothing because it is scalable to large-scale neural networks and applicable to any classifier. We derive a tight robustness in $\ell_2$ norm for top-$k$ predictions when using randomized smoothing with Gaussian noise. We find that generalizing the certified robustness from top-1 to top-$k$ predictions faces significant technical challenges. We also empirically evaluate our method on CIFAR10 and ImageNet. For example, our method can obtain an ImageNet classifier with a certified top-5 accuracy of 62.8\% when the $\ell_2$-norms of the adversarial perturbations are less than 0.5 (=127/255). Our code is publicly available at: \url{https://github.com/jjy1994/Certify_Topk}. | We study the certified robustness for top-k predictions via randomized smoothing under Gaussian noise and derive a tight robustness bound in L_2 norm. |
Recent work has shown increased interest in using the Variational Autoencoder (VAE) framework to discover interpretable representations of data in an unsupervised way. These methods have focussed largely on modifying the variational cost function to achieve this goal. However, we show that methods like beta-VAE simplify the tendency of variational inference to underfit causing pathological over-pruning and over-orthogonalization of learned components. In this paper we take a complementary approach: to modify the probabilistic model to encourage structured latent variable representations to be discovered. Specifically, the standard VAE probabilistic model is unidentifiable: the likelihood of the parameters is invariant under rotations of the latent space. This means there is no pressure to identify each true factor of variation with a latent variable.
We therefore employ a rich prior distribution, akin to the ICA model, that breaks the rotational symmetry.
Extensive quantitative and qualitative experiments demonstrate that the proposed prior mitigates the trade-off introduced by modified cost functions like beta-VAE and TCVAE between reconstruction loss and disentanglement. The proposed prior allows to improve these approaches with respect to both disentanglement and reconstruction quality significantly over the state of the art. | We present structured priors for unsupervised learning of disentangled representations in VAEs that significantly mitigate the trade-off between disentanglement and reconstruction loss. |
Due to the success of residual networks (resnets) and related architectures, shortcut connections have quickly become standard tools for building convolutional neural networks. The explanations in the literature for the apparent effectiveness of shortcuts are varied and often contradictory. We hypothesize that shortcuts work primarily because they act as linear counterparts to nonlinear layers. We test this hypothesis by using several variations on the standard residual block, with different types of linear connections, to build small (100k--1.2M parameter) image classification networks. Our experiments show that other kinds of linear connections can be even more effective than the identity shortcuts. Our results also suggest that the best type of linear connection for a given application may depend on both network width and depth. | We generalize residual blocks to tandem blocks, which use arbitrary linear maps instead of shortcuts, and improve performance over ResNets. |
Adam-typed optimizers, as a class of adaptive moment estimation methods with the exponential moving average scheme, have been successfully used in many applications of deep learning. Such methods are appealing for capability on large-scale sparse datasets. On top of that, they are computationally efficient and insensitive to the hyper-parameter settings. In this paper, we present a new framework for adapting Adam-typed methods, namely AdamT. Instead of applying a simple exponential weighted average, AdamT also includes the trend information when updating the parameters with the adaptive step size and gradients. The newly added term is expected to efficiently capture the non-horizontal moving patterns on the cost surface, and thus converge more rapidly. We show empirically the importance of the trend component, where AdamT outperforms the conventional Adam method constantly in both convex and non-convex settings. | We present a new framework for adapting Adam-typed methods, namely AdamT, to include the trend information when updating the parameters with the adaptive step size and gradients. |
As machine learning methods see greater adoption and implementation in high stakes applications such
as medical image diagnosis, the need for model interpretability and explanation has become more
critical. Classical approaches that assess feature importance (eg saliency maps) do not explain how and why a particular region of an image is relevant to the prediction. We propose
a method that explains the outcome of a classification black-box by gradually exaggerating
the semantic effect of a given class. Given a query input to a classifier, our method produces a
progressive set of plausible variations of that query, which gradually change the posterior probability
from its original class to its negation. These counter-factually generated samples preserve features
unrelated to the classification decision, such that a user can employ our method as a ``tuning knob'' to traverse a data manifold while crossing the decision boundary. Our method is model agnostic and only requires the output value and gradient of the predictor with respect to its input. | A method to explain a classifier, by generating visual perturbation of an image by exaggerating or diminishing the semantic features that the classifier associates with a target label. |
We study the problem of explaining a rich class of behavioral properties of deep neural networks. Our influence-directed explanations approach this problem by peering inside the network to identify neurons with high influence on the property of interest using an axiomatically justified influence measure, and then providing an interpretation for the concepts these neurons represent. We evaluate our approach by training convolutional neural networks on Pubfig, ImageNet, and Diabetic Retinopathy datasets. Our evaluation demonstrates that influence-directed explanations (1) localize features used by the network, (2) isolate features distinguishing related instances, (3) help extract the essence of what the network learned about the class, and (4) assist in debugging misclassifications.
| We present an influence-directed approach to constructing explanations for the behavior of deep convolutional networks, and show how it can be used to answer a broad set of questions that could not be addressed by prior work. |
Standard deep learning systems require thousands or millions of examples to learn a concept, and cannot integrate new concepts easily. By contrast, humans have an incredible ability to do one-shot or few-shot learning. For instance, from just hearing a word used in a sentence, humans can infer a great deal about it, by leveraging what the syntax and semantics of the surrounding words tells us. Here, we draw inspiration from this to highlight a simple technique by which deep recurrent networks can similarly exploit their prior knowledge to learn a useful representation for a new word from little data. This could make natural language processing systems much more flexible, by allowing them to learn continually from the new words they encounter. | We highlight a technique by which natural language processing systems can learn a new word from context, allowing them to be much more flexible. |
Recent research developing neural network architectures with external memory have often used the benchmark bAbI question and answering dataset which provides a challenging number of tasks requiring reasoning. Here we employed a classic associative inference task from the human neuroscience literature in order to more carefully probe the reasoning capacity of existing memory-augmented architectures. This task is thought to capture the essence of reasoning -- the appreciation of distant relationships among elements distributed across multiple facts or memories. Surprisingly, we found that current architectures struggle to reason over long distance associations. Similar results were obtained on a more complex task involving finding the shortest path between nodes in a path. We therefore developed a novel architecture, MEMO, endowed with the capacity to reason over longer distances. This was accomplished with the addition of two novel components. First, it introduces a separation between memories/facts stored in external memory and the items that comprise these facts in external memory. Second, it makes use of an adaptive retrieval mechanism, allowing a variable number of ‘memory hops’ before the answer is produced. MEMO is capable of solving our novel reasoning tasks, as well as all 20 tasks in bAbI. | A memory architecture that support inferential reasoning. |
Depthwise separable convolutions reduce the number of parameters and computation used in convolutional operations while increasing representational efficiency.
They have been shown to be successful in image classification models, both in obtaining better models than previously possible for a given parameter count (the Xception architecture) and considerably reducing the number of parameters required to perform at a given level (the MobileNets family of architectures). Recently, convolutional sequence-to-sequence networks have been applied to machine translation tasks with good results. In this work, we study how depthwise separable convolutions can be applied to neural machine translation. We introduce a new architecture inspired by Xception and ByteNet, called SliceNet, which enables a significant reduction of the parameter count and amount of computation needed to obtain results like ByteNet, and, with a similar parameter count, achieves better results.
In addition to showing that depthwise separable convolutions perform well for machine translation, we investigate the architectural changes that they enable: we observe that thanks to depthwise separability, we can increase the length of convolution windows, removing the need for filter dilation. We also introduce a new super-separable convolution operation that further reduces the number of parameters and computational cost of the models. | Depthwise separable convolutions improve neural machine translation: the more separable the better. |
Interpreting generative adversarial network (GAN) training as approximate divergence minimization has been
theoretically insightful, has spurred discussion, and has lead to theoretically and practically interesting
extensions such as f-GANs and Wasserstein GANs. For both classic GANs and f-GANs, there is an original variant of training and a "non-saturating" variant which uses an alternative form of generator gradient. The original variant is theoretically easier to study, but for GANs the alternative variant performs better in practice. The non-saturating scheme is often regarded as a simple modification to deal with optimization issues, but we show that in fact the non-saturating scheme for GANs is effectively optimizing a reverse KL-like f-divergence. We also develop a number of theoretical tools to help compare and classify f-divergences. We hope these results may help to clarify some of the theoretical discussion surrounding the divergence minimization view of GAN training. | Non-saturating GAN training effectively minimizes a reverse KL-like f-divergence. |
We introduce a novel method for converting text data into abstract image representations, which allows image-based processing techniques (e.g. image classification networks) to be applied to text-based comparison problems. We apply the technique to entity disambiguation of inventor names in US patents. The method involves converting text from each pairwise comparison between two inventor name records into a 2D RGB (stacked) image representation. We then train an image classification neural network to discriminate between such pairwise comparison images, and use the trained network to label each pair of records as either matched (same inventor) or non-matched (different inventors), obtaining highly accurate results (F1: 99.09%, precision: 99.41%, recall: 98.76%). Our new text-to-image representation method could potentially be used more broadly for other NLP comparison problems, such as disambiguation of academic publications, or for problems that require simultaneous classification of both text and images. | We introduce a novel text representation method which enables image classifiers to be applied to text classification problems, and apply the method to inventor name disambiguation. |
We propose a novel algorithm, Difference-Seeking Generative Adversarial Network (DSGAN), developed from traditional GAN. DSGAN considers the scenario that the training samples of target distribution, $p_{t}$, are difficult to collect.
Suppose there are two distributions $p_{\bar{d}}$ and $p_{d}$ such that the density of the target distribution can be the differences between the densities of $p_{\bar{d}}$ and $p_{d}$. We show how to learn the target distribution $p_{t}$ only via samples from $p_{d}$ and $p_{\bar{d}}$ (relatively easy to obtain).
DSGAN has the flexibility to produce samples from various target distributions (e.g. the out-of-distribution). Two key applications, semi-supervised learning and adversarial training, are taken as examples to validate the effectiveness of DSGAN. We also provide theoretical analyses about the convergence of DSGAN. | We proposed "Difference-Seeking Generative Adversarial Network" (DSGAN) model to learn the target distribution which is hard to collect training data. |
Recently, Generative Adversarial Network (GAN) and numbers of its variants have been widely used to solve the image-to-image translation problem and achieved extraordinary results in both a supervised and unsupervised manner. However, most GAN-based methods suffer from the imbalance problem between the generator and discriminator in practice. Namely, the relative model capacities of the generator and discriminator do not match, leading to mode collapse and/or diminished gradients. To tackle this problem, we propose a GuideGAN based on attention mechanism. More specifically, we arm the discriminator with an attention mechanism so not only it estimates the probability that its input is real, but also does it create an attention map that highlights the critical features for such prediction. This attention map then assists the generator to produce more plausible and realistic images. We extensively evaluate the proposed GuideGAN framework on a number of image transfer tasks. Both qualitative results and quantitative comparison demonstrate the superiority of our proposed approach. | A general method that improves the image translation performance of GAN framework by using an attention embedded discriminator |
The problem of verifying whether a textual hypothesis holds based on the given evidence, also known as fact verification, plays an important role in the study of natural language understanding and semantic representation. However, existing studies are mainly restricted to dealing with unstructured evidence (e.g., natural language sentences and documents, news, etc), while verification under structured evidence, such as tables, graphs, and databases, remains unexplored. This paper specifically aims to study the fact verification given semi-structured data as evidence. To this end, we construct a large-scale dataset called TabFact with 16k Wikipedia tables as the evidence for 118k human-annotated natural language statements, which are labeled as either ENTAILED or REFUTED. TabFact is challenging since it involves both soft linguistic reasoning and hard symbolic reasoning. To address these reasoning challenges, we design two different models: Table-BERT and Latent Program Algorithm (LPA). Table-BERT leverages the state-of-the-art pre-trained language model to encode the linearized tables and statements into continuous vectors for verification. LPA parses statements into LISP-like programs and executes them against the tables to obtain the returned binary value for verification. Both methods achieve similar accuracy but still lag far behind human performance. We also perform a comprehensive analysis to demonstrate great future opportunities. | We propose a new dataset to investigate the entailment problem under semi-structured table as premise |
This work presents a two-stage neural architecture for learning and refining structural correspondences between graphs. First, we use localized node embeddings computed by a graph neural network to obtain an initial ranking of soft correspondences between nodes. Secondly, we employ synchronous message passing networks to iteratively re-rank the soft correspondences to reach a matching consensus in local neighborhoods between graphs. We show, theoretically and empirically, that our message passing scheme computes a well-founded measure of consensus for corresponding neighborhoods, which is then used to guide the iterative re-ranking process. Our purely local and sparsity-aware architecture scales well to large, real-world inputs while still being able to recover global correspondences consistently. We demonstrate the practical effectiveness of our method on real-world tasks from the fields of computer vision and entity alignment between knowledge graphs, on which we improve upon the current state-of-the-art. | We develop a deep graph matching architecture which refines initial correspondences in order to reach neighborhood consensus. |
This paper extends the proof of density of neural networks in the space of continuous (or even measurable) functions on Euclidean spaces to functions on compact sets of probability measures.
By doing so the work parallels a more then a decade old results on mean-map embedding of probability measures in reproducing kernel Hilbert spaces.
The work has wide practical consequences for multi-instance learning, where it theoretically justifies some recently proposed constructions.
The result is then extended to Cartesian products, yielding universal approximation theorem for tree-structured domains, which naturally occur in data-exchange formats like JSON, XML, YAML, AVRO, and ProtoBuffer. This has important practical implications, as it enables to automatically create an architecture of neural networks for processing structured data (AutoML paradigms), as demonstrated by an accompanied library for JSON format. | This paper extends the proof of density of neural networks in the space of continuous (or even measurable) functions on Euclidean spaces to functions on compact sets of probability measures. |
Interactions such as double negation in sentences and scene interactions in images are common forms of complex dependencies captured by state-of-the-art machine learning models. We propose Mahé, a novel approach to provide Model-Agnostic Hierarchical Explanations of how powerful machine learning models, such as deep neural networks, capture these interactions as either dependent on or free of the context of data instances. Specifically, Mahé provides context-dependent explanations by a novel local interpretation algorithm that effectively captures any-order interactions, and obtains context-free explanations through generalizing context-dependent interactions to explain global behaviors. Experimental results show that Mahé obtains improved local interaction interpretations over state-of-the-art methods and successfully provides explanations of interactions that are context-free. | A new framework for context-dependent and context-free explanations of predictions |
To realize the promise of ubiquitous embedded deep network inference, it is essential to seek limits of energy and area efficiency. To this end, low-precision networks offer tremendous promise because both energy and area scale down quadratically with the reduction in precision. Here, for the first time, we demonstrate ResNet-18, ResNet-34, ResNet-50, ResNet-152, Inception-v3, densenet-161, and VGG-16bn networks on the ImageNet classification benchmark that, at 8-bit precision exceed the accuracy of the full-precision baseline networks after one epoch of finetuning, thereby leveraging the availability of pretrained models.
We also demonstrate ResNet-18, ResNet-34, and ResNet-50 4-bit models that match the accuracy of the full-precision baseline networks -- the highest scores to date. Surprisingly, the weights of the low-precision networks are very close (in cosine similarity) to the weights of the corresponding baseline networks, making training from scratch unnecessary.
We find that gradient noise due to quantization during training increases with reduced precision, and seek ways to overcome this noise. The number of iterations required by stochastic gradient descent to achieve a given training error is related to the square of (a) the distance of the initial solution from the final plus (b) the maximum variance of the gradient estimates. By drawing inspiration from this observation, we (a) reduce solution distance by starting with pretrained fp32 precision baseline networks and fine-tuning, and (b) combat noise introduced by quantizing weights and activations during training, by using larger batches along with matched learning rate annealing. Sensitivity analysis indicates that these techniques, coupled with proper activation function range calibration, offer a promising heuristic to discover low-precision networks, if they exist, close to fp32 precision baseline networks.
| Finetuning after quantization matches or exceeds full-precision state-of-the-art networks at both 8- and 4-bit quantization. |
Analysis methods which enable us to better understand the
representations and functioning of neural models of language are
increasingly needed as deep learning becomes the dominant approach
in NLP. Here we present two methods based on Representational
Similarity Analysis (RSA) and Tree Kernels (TK) which allow us to
directly quantify how strongly the information encoded in neural
activation patterns corresponds to information represented by
symbolic structures such as syntax trees. We first validate our
methods on the case of a simple synthetic language for arithmetic
expressions with clearly defined syntax and semantics, and show that
they exhibit the expected pattern of results. We then apply our methods to
correlate neural representations of English sentences with their
constituency parse trees. | Two methods based on Representational Similarity Analysis (RSA) and Tree Kernels (TK) which directly quantify how strongly information encoded in neural activation patterns corresponds to information represented by symbolic structures. |
Supervised deep learning requires a large amount of training samples with annotations (e.g. label class for classification task, pixel- or voxel-wised label map for segmentation tasks), which are expensive and time-consuming to obtain. During the training of a deep neural network, the annotated samples are fed into the network in a mini-batch way, where they are often regarded of equal importance. However, some of the samples may become less informative during training, as the magnitude of the gradient start to vanish for these samples. In the meantime, other samples of higher utility or hardness may be more demanded for the training process to proceed and require more exploitation. To address the challenges of expensive annotations and loss of sample informativeness, here we propose a novel training framework which adaptively selects informative samples that are fed to the training process. The adaptive selection or sampling is performed based on a hardness-aware strategy in the latent space constructed by a generative model. To evaluate the proposed training framework, we perform experiments on three different datasets, including MNIST and CIFAR-10 for image classification task and a medical image dataset IVUS for biophysical simulation task. On all three datasets, the proposed framework outperforms a random sampling method, which demonstrates the effectiveness of our framework. | This paper introduces a framework for data-efficient representation learning by adaptive sampling in latent space. |
Existing methods for AI-generated artworks still struggle with generating high-quality stylized content, where high-level semantics are preserved, or separating fine-grained styles from various artists. We propose a novel Generative Adversarial Disentanglement Network which can disentangle two complementary factors of variations when only one of them is labelled in general, and fully decompose complex anime illustrations into style and content in particular. Training such model is challenging, since given a style, various content data may exist but not the other way round. Our approach is divided into two stages, one that encodes an input image into a style independent content, and one based on a dual-conditional generator. We demonstrate the ability to generate high-fidelity anime portraits with a fixed content and a large variety of styles from over a thousand artists, and vice versa, using a single end-to-end network and with applications in style transfer. We show this unique capability as well as superior output to the current state-of-the-art. | An adversarial training-based method for disentangling two complementary sets of variations in a dataset where only one of them is labelled, tested on style vs. content in anime illustrations. |
Recent research has shown that CNNs are often overly sensitive to high-frequency textural patterns. Inspired by the intuition that humans are more sensitive to the lower-frequency (larger-scale) patterns we design a regularization scheme that penalizes large differences between adjacent components within each convolutional kernel. We apply our regularization onto several popular training methods, demonstrating that the models with the proposed smooth kernels enjoy improved adversarial robustness. Further, building on recent work establishing connections between adversarial robustness and interpretability, we show that our method appears to give more perceptually-aligned gradients. | We introduce a smoothness regularization for convolutional kernels of CNN that can help improve adversarial robustness and lead to perceptually-aligned gradients |
Despite an ever growing literature on reinforcement learning algorithms and applications, much less is known about their statistical inference. In this paper, we investigate the large-sample behaviors of the Q-value estimates with closed-form characterizations of the asymptotic variances. This allows us to efficiently construct confidence regions for Q-value and optimal value functions, and to develop policies to minimize their estimation errors. This also leads to a policy exploration strategy that relies on estimating the relative discrepancies among the Q estimates. Numerical experiments show superior performances of our exploration strategy than other benchmark approaches. | We investigate the large-sample behaviors of the Q-value estimates and proposed an efficient exploration strategy that relies on estimating the relative discrepancies among the Q estimates. |
Entailment vectors are a principled way to encode in a vector what information is known and what is unknown. They are designed to model relations where one vector should include all the information in another vector, called entailment. This paper investigates the unsupervised learning of entailment vectors for the semantics of words. Using simple entailment-based models of the semantics of words in text (distributional semantics), we induce entailment-vector word embeddings which outperform the best previous results for predicting entailment between words, in unsupervised and semi-supervised experiments on hyponymy.
| We train word embeddings based on entailment instead of similarity, successfully predicting lexical entailment. |
We describe a simple scheme that allows an agent to learn about its environment in an unsupervised manner. Our scheme pits two versions of the same agent, Alice and Bob, against one another. Alice proposes a task for Bob to complete; and then Bob attempts to complete the task. In this work we will focus on two kinds of environments: (nearly) reversible environments and environments that can be reset. Alice will "propose" the task by doing a sequence of actions and then Bob must undo or repeat them, respectively. Via an appropriate reward structure, Alice and Bob automatically generate a curriculum of exploration, enabling unsupervised training of the agent. When Bob is deployed on an RL task within the environment, this unsupervised training reduces the number of supervised episodes needed to learn, and in some cases converges to a higher reward. | Unsupervised learning for reinforcement learning using an automatic curriculum of self-play |
Many real-world data sets are represented as graphs, such as citation links, social media, and biological interaction. The volatile graph structure makes it non-trivial to employ convolutional neural networks (CNN's) for graph data processing. Recently, graph attention network (GAT) has proven a promising attempt by combining graph neural networks with attention mechanism, so as to achieve massage passing in graphs with arbitrary structures. However, the attention in GAT is computed mainly based on the similarity between the node content, while the structures of the graph remains largely unemployed (except in masking the attention out of one-hop neighbors). In this paper, we propose an `````````````````````````````"ADaptive Structural Fingerprint" (ADSF) model to fully exploit both topological details of the graph and content features of the nodes. The key idea is to contextualize each node with a weighted, learnable receptive field encoding rich and diverse local graph structures. By doing this, structural interactions between the nodes can be inferred accurately, thus improving subsequent attention layer as well as the convergence of learning. Furthermore, our model provides a useful platform for different subspaces of node features and various scales of graph structures to ``cross-talk'' with each other through the learning of multi-head attention, being particularly useful in handling complex real-world data. Encouraging performance is observed on a number of benchmark data sets in node classification. | Exploiting rich strucural details in graph-structued data via adaptive "strucutral fingerprints'' |
Informed and robust decision making in the face of uncertainty is critical for robots that perform physical tasks alongside people. We formulate this as a Bayesian Reinforcement Learning problem over latent Markov Decision Processes (MDPs). While Bayes-optimality is theoretically the gold standard, existing algorithms do not scale well to continuous state and action spaces. We propose a scalable solution that builds on the following insight: in the absence of uncertainty, each latent MDP is easier to solve. We split the challenge into two simpler components. First, we obtain an ensemble of clairvoyant experts and fuse their advice to compute a baseline policy. Second, we train a Bayesian residual policy to improve upon the ensemble's recommendation and learn to reduce uncertainty. Our algorithm, Bayesian Residual Policy Optimization (BRPO), imports the scalability of policy gradient methods as well as the initialization from prior models. BRPO significantly improves the ensemble of experts and drastically outperforms existing adaptive RL methods. | We propose a scalable Bayesian Reinforcement Learning algorithm that learns a Bayesian correction over an ensemble of clairvoyant experts to solve problems with complex latent rewards and dynamics. |
One of the mysteries in the success of neural networks is randomly initialized first order methods like gradient descent can achieve zero training loss even though the objective function is non-convex and non-smooth. This paper demystifies this surprising phenomenon for two-layer fully connected ReLU activated neural networks. For an $m$ hidden node shallow neural network with ReLU activation and $n$ training data, we show as long as $m$ is large enough and no two inputs are parallel, randomly initialized gradient descent converges to a globally optimal solution at a linear convergence rate for the quadratic loss function.
Our analysis relies on the following observation: over-parameterization and random initialization jointly restrict every weight vector to be close to its initialization for all iterations, which allows us to exploit a strong convexity-like property to show that gradient descent converges at a global linear rate to the global optimum. We believe these insights are also useful in analyzing deep models and other first order methods. | We prove gradient descent achieves zero training loss with a linear rate on over-parameterized neural networks. |
For many applications, in particular in natural science, the task is to
determine hidden system parameters from a set of measurements. Often,
the forward process from parameter- to measurement-space is well-defined,
whereas the inverse problem is ambiguous: multiple parameter sets can
result in the same measurement. To fully characterize this ambiguity, the full
posterior parameter distribution, conditioned on an observed measurement,
has to be determined. We argue that a particular class of neural networks
is well suited for this task – so-called Invertible Neural Networks (INNs).
Unlike classical neural networks, which attempt to solve the ambiguous
inverse problem directly, INNs focus on learning the forward process, using
additional latent output variables to capture the information otherwise
lost. Due to invertibility, a model of the corresponding inverse process is
learned implicitly. Given a specific measurement and the distribution of
the latent variables, the inverse pass of the INN provides the full posterior
over parameter space. We prove theoretically and verify experimentally, on
artificial data and real-world problems from medicine and astrophysics, that
INNs are a powerful analysis tool to find multi-modalities in parameter space,
uncover parameter correlations, and identify unrecoverable parameters. | To analyze inverse problems with Invertible Neural Networks |
Decisions made by machine learning systems have increasing influence on the world. Yet it is common for machine learning algorithms to assume that no such influence exists. An example is the use of the i.i.d. assumption in online learning for applications such as content recommendation, where the (choice of) content displayed can change users' perceptions and preferences, or even drive them away, causing a shift in the distribution of users. Generally speaking, it is possible for an algorithm to change the distribution of its own inputs. We introduce the term self-induced distributional shift (SIDS) to describe this phenomenon. A large body of work in reinforcement learning and causal machine learning aims to deal with distributional shift caused by deploying learning systems previously trained offline. Our goal is similar, but distinct: we point out that changes to the learning algorithm, such as the introduction of meta-learning, can reveal hidden incentives for distributional shift (HIDS), and aim to diagnose and prevent problems associated with hidden incentives. We design a simple environment as a "unit test" for HIDS, as well as a content recommendation environment which allows us to disentangle different types of SIDS. We demonstrate the potential for HIDS to cause unexpected or undesirable behavior in these environments, and propose and test a mitigation strategy. | Performance metrics are incomplete specifications; the ends don't always justify the means. |
In one-class-learning tasks, only the normal case can be modeled with data, whereas the variation of all possible anomalies is too large to be described sufficiently by samples. Thus, due to the lack of representative data, the wide-spread discriminative approaches cannot cover such learning tasks, and rather generative models, which attempt to learn the input density of the normal cases, are used. However, generative models suffer from a large input dimensionality (as in images) and are typically inefficient learners. We propose to learn the data distribution more efficiently with a multi-hypotheses autoencoder. Moreover, the model is criticized by a discriminator, which prevents artificial data modes not supported by data, and which enforces diversity across hypotheses. This consistency-based anomaly detection (ConAD) framework allows the reliable identification of outof- distribution samples. For anomaly detection on CIFAR-10, it yields up to 3.9% points improvement over previously reported results. On a real anomaly detection task, the approach reduces the error of the baseline models from 6.8% to 1.5%. | We propose an anomaly-detection approach that combines modeling the foreground class via multiple local densities with adversarial training. |
Generative Adversarial Networks (GAN) can achieve promising performance on learning complex data distributions on different types of data. In this paper, we first show that a straightforward extension of an existing GAN algorithm is not applicable to point clouds, because the constraint required for discriminators is undefined for set data. We propose a two fold modification to a GAN algorithm to be able to generate point clouds (PC-GAN). First, we combine ideas from hierarchical Bayesian modeling and implicit generative models by learning a hierarchical and interpretable sampling process. A key component of our method is that we train a posterior inference network for the hidden variables. Second, PC-GAN defines a generic framework that can incorporate many existing GAN algorithms. We further propose a sandwiching objective, which results in a tighter Wasserstein distance estimate than the commonly used dual form in WGAN. We validate our claims on the ModelNet40 benchmark dataset and observe that PC- GAN trained by the sandwiching objective achieves better results on test data than existing methods. We also conduct studies on several tasks, including generalization on unseen point clouds, latent space interpolation, classification, and image to point clouds transformation, to demonstrate the versatility of the proposed PC-GAN algorithm. | We propose a GAN variant which learns to generate point clouds. Different studies have been explores, including tighter Wasserstein distance estimate, conditional generation, generalization to unseen point clouds and image to point cloud. |
Existing attention mechanisms, are mostly item-based in that a model is trained to attend to individual items in a collection (the memory) where each item has a predefined, fixed granularity, e.g., a character or a word. Intuitively, an area in the memory consisting of multiple items can be worth attending to as a whole. We propose area attention: a way to attend to an area of the memory, where each area contains a group of items that are either spatially adjacent when the memory has a 2-dimensional structure, such as images, or temporally adjacent for 1-dimensional memory, such as natural language sentences. Importantly, the size of an area, i.e., the number of items in an area or the level of aggregation, is dynamically determined via learning, which can vary depending on the learned coherence of the adjacent items. By giving the model the option to attend to an area of items, instead of only individual items, a model can attend to information with varying granularity. Area attention can work along multi-head attention for attending to multiple areas in the memory. We evaluate area attention on two tasks: neural machine translation (both character and token-level) and image captioning, and improve upon strong (state-of-the-art) baselines in all the cases. These improvements are obtainable with a basic form of area attention that is parameter free. In addition to proposing the novel concept of area attention, we contribute an efficient way for computing it by leveraging the technique of summed area tables. | The paper presents a novel approach for attentional mechanisms that can benefit a range of tasks such as machine translation and image captioning. |
We identify a phenomenon, which we refer to as *multi-model forgetting*, that occurs when sequentially training multiple deep networks with partially-shared parameters; the performance of previously-trained models degrades as one optimizes a subsequent one, due to the overwriting of shared parameters. To overcome this, we introduce a statistically-justified weight plasticity loss that regularizes the learning of a model's shared parameters according to their importance for the previous models, and demonstrate its effectiveness when training two models sequentially and for neural architecture search. Adding weight plasticity in neural architecture search preserves the best models to the end of the search and yields improved results in both natural language processing and computer vision tasks. | We identify a phenomenon, neural brainwashing, and introduce a statistically-justified weight plasticity loss to overcome this. |
Revealing latent structure in data is an active field of research, having introduced exciting technologies such as variational autoencoders and adversarial networks, and is essential to push machine learning towards unsupervised knowledge discovery. However, a major challenge is the lack of suitable benchmarks for an objective and quantitative evaluation of learned representations. To address this issue we introduce Morpho-MNIST, a framework that aims to answer: "to what extent has my model learned to represent specific factors of variation in the data?" We extend the popular MNIST dataset by adding a morphometric analysis enabling quantitative comparison of trained models, identification of the roles of latent variables, and characterisation of sample diversity. We further propose a set of quantifiable perturbations to assess the performance of unsupervised and supervised methods on challenging tasks such as outlier detection and domain adaptation. | This paper introduces Morpho-MNIST, a collection of shape metrics and perturbations, in a step towards quantitative evaluation of representation learning. |
Exploration in environments with sparse rewards is a key challenge for reinforcement learning. How do we design agents with generic inductive biases so that they can explore in a consistent manner instead of just using local exploration schemes like epsilon-greedy? We propose an unsupervised reinforcement learning agent which learns a discrete pixel grouping model that preserves spatial geometry of the sensors and implicitly of the environment as well. We use this representation to derive geometric intrinsic reward functions, like centroid coordinates and area, and learn policies to control each one of them with off-policy learning. These policies form a basis set of behaviors (options) which allows us explore in a consistent way and use them in a hierarchical reinforcement learning setup to solve for extrinsically defined rewards. We show that our approach can scale to a variety of domains with competitive performance, including navigation in 3D environments and Atari games with sparse rewards. | structured exploration in deep reinforcement learning via unsupervised visual abstraction discovery and control |
Combinatorial optimization is a common theme in computer science. While in general such problems are NP-Hard, from a practical point of view, locally optimal solutions can be useful. In some combinatorial problems however, it can be hard to define meaningful solution neighborhoods that connect large portions of the search space, thus hindering methods that search this space directly. We suggest to circumvent such cases by utilizing a policy gradient algorithm that transforms the problem to the continuous domain, and to optimize a new surrogate objective that renders the former as generic stochastic optimizer. This is achieved by producing a surrogate objective whose distribution is fixed and predetermined, thus removing the need to fine-tune various hyper-parameters in a case by case manner. Since we are interested in methods which can successfully recover locally optimal solutions, we use the problem of finding locally maximal cliques as a challenging experimental benchmark, and we report results on a large dataset of graphs that is designed to test clique finding algorithms. Notably, we show in this benchmark that fixing the distribution of the surrogate is key to consistently recovering locally optimal solutions, and that our surrogate objective leads to an algorithm that outperforms other methods we have tested in a number of measures. | A new policy gradient algorithm designed to approach black-box combinatorial optimization problems. The algorithm relies only on function evaluations, and returns locally optimal solutions with high probability. |
Deterministic neural networks (NNs) are increasingly being deployed in safety critical domains, where calibrated, robust and efficient measures of uncertainty are crucial. While it is possible to train regression networks to output the parameters of a probability distribution by maximizing a Gaussian likelihood function, the resulting model remains oblivious to the underlying confidence of its predictions. In this paper, we propose a novel method for training deterministic NNs to not only estimate the desired target but also the associated evidence in support of that target. We accomplish this by placing evidential priors over our original Gaussian likelihood function and training our NN to infer the hyperparameters of our evidential distribution. We impose priors during training such that the model is penalized when its predicted evidence is not aligned with the correct output. Thus the model estimates not only the probabilistic mean and variance of our target but also the underlying uncertainty associated with each of those parameters. We observe that our evidential regression method learns well-calibrated measures of uncertainty on various benchmarks, scales to complex computer vision tasks, and is robust to adversarial input perturbations.
| Fast, calibrated uncertainty estimation for neural networks without sampling |
The Lottery Ticket Hypothesis from Frankle & Carbin (2019) conjectures that, for typically-sized neural networks, it is possible to find small sub-networks which train faster and yield superior performance than their original counterparts. The proposed algorithm to search for such sub-networks (winning tickets), Iterative Magnitude Pruning (IMP), consistently finds sub-networks with 90-95% less parameters which indeed train faster and better than the overparameterized models they were extracted from, creating potential applications to problems such as transfer learning.
In this paper, we propose a new algorithm to search for winning tickets, Continuous Sparsification, which continuously removes parameters from a network during training, and learns the sub-network's structure with gradient-based methods instead of relying on pruning strategies. We show empirically that our method is capable of finding tickets that outperforms the ones learned by Iterative Magnitude Pruning, and at the same time providing up to 5 times faster search, when measured in number of training epochs. | We propose a new algorithm that quickly finds winning tickets in neural networks. |
In most practical settings and theoretical analyses, one assumes that a model can be trained until convergence. However, the growing complexity of machine learning datasets and models may violate such assumptions. Indeed, current approaches for hyper-parameter tuning and neural architecture search tend to be limited by practical resource constraints. Therefore, we introduce a formal setting for studying training under the non-asymptotic, resource-constrained regime, i.e., budgeted training. We analyze the following problem: "given a dataset, algorithm, and fixed resource budget, what is the best achievable performance?" We focus on the number of optimization iterations as the representative resource. Under such a setting, we show that it is critical to adjust the learning rate schedule according to the given budget. Among budget-aware learning schedules, we find simple linear decay to be both robust and high-performing. We support our claim through extensive experiments with state-of-the-art models on ImageNet (image classification), Kinetics (video classification), MS COCO (object detection and instance segmentation), and Cityscapes (semantic segmentation). We also analyze our results and find that the key to a good schedule is budgeted convergence, a phenomenon whereby the gradient vanishes at the end of each allowed budget. We also revisit existing approaches for fast convergence and show that budget-aware learning schedules readily outperform such approaches under (the practical but under-explored) budgeted training setting. | Introduce a formal setting for budgeted training and propose a budget-aware linear learning rate schedule |
We present a new approach for efficient exploration which leverages a low-dimensional encoding of the environment learned with a combination of model-based and model-free objectives. Our approach uses intrinsic rewards that are based on a weighted distance of nearest neighbors in the low dimensional representational space to gauge novelty.
We then leverage these intrinsic rewards for sample-efficient exploration with planning routines in representational space.
One key element of our approach is that we perform more gradient steps in-between every environment step in order to ensure the model accuracy. We test our approach on a number of maze tasks, as well as a control problem and show that our exploration approach is more sample-efficient compared to strong baselines. | We conduct exploration using intrinsic rewards that are based on a weighted distance of nearest neighbors in representational space. |
Neural networks are vulnerable to small adversarial perturbations. While existing literature largely focused on the vulnerability of learned models, we demonstrate an intriguing phenomenon that adversarial robustness, unlike clean accuracy, is sensitive to the input data distribution. Even a semantics-preserving transformations on the input data distribution can cause a significantly different robustness for the adversarially trained model that is both trained and evaluated on the new distribution. We show this by constructing semantically- identical variants for MNIST and CIFAR10 respectively, and show that standardly trained models achieve similar clean accuracies on them, but adversarially trained models achieve significantly different robustness accuracies. This counter-intuitive phenomenon indicates that input data distribution alone can affect the adversarial robustness of trained neural networks, not necessarily the tasks themselves. Lastly, we discuss the practical implications on evaluating adversarial robustness, and make initial attempts to understand this complex phenomenon. | Robustness performance of PGD trained models are sensitive to semantics-preserving transformation of image datasets, which implies the trickiness of evaluation of robust learning algorithms in practice. |
Sample inefficiency is a long-lasting problem in reinforcement learning (RL). The state-of-the-art uses action value function to derive policy while it usually involves an extensive search over the state-action space and unstable optimization. Towards the sample-efficient RL, we propose ranking policy gradient (RPG), a policy gradient method that learns the optimal rank of a set of discrete actions. To accelerate the learning of policy gradient methods, we establish the equivalence between maximizing the lower bound of return and imitating a near-optimal policy without accessing any oracles. These results lead to a general off-policy learning framework, which preserves the optimality, reduces variance, and improves the sample-efficiency. We conduct extensive experiments showing that when consolidating with the off-policy learning framework, RPG substantially reduces the sample complexity, comparing to the state-of-the-art. | We propose ranking policy gradient that learns the optimal rank of actions to maximize return. We propose a general off-policy learning framework with the properties of optimality preserving, variance reduction, and sample-efficiency. |
We introduce MultiGrain, a neural network architecture that generates compact image embedding vectors that solve multiple tasks of different granularity: class, instance, and copy recognition. MultiGrain is trained jointly for classification by optimizing the cross-entropy loss and for instance/copy recognition by optimizing a self-supervised ranking loss. The self-supervised loss only uses data augmentation and thus does not require additional labels. Remarkably, the unified embeddings are not only much more compact than using several specialized embeddings, but they also have the same or better accuracy. When fed to a linear classifier, MultiGrain using ResNet-50 achieves 79.4% top-1 accuracy on ImageNet, a +1.8% absolute improvement over the the current state-of-the-art AutoAugment method. The same embeddings perform on par with state-of-the-art instance retrieval with images of moderate resolution. An ablation study shows that our approach benefits from the self-supervision, the pooling method and the mini-batches with repeated augmentations of the same image.
| Combining classification and image retrieval in a neural network architecture, we obtain an improvement for both tasks. |
In this paper, we investigate mapping the hyponymy relation of
wordnet to feature vectors.
We aim to model lexical knowledge in such a way that it can be used as
input in generic machine-learning models, such as phrase entailment
predictors.
We propose two models. The first one leverages an existing mapping of
words to feature vectors (fasttext), and attempts to classify
such vectors as within or outside of each class. The second model is fully supervised,
using solely wordnet as a ground truth. It maps each concept to an
interval or a disjunction thereof.
On the first model, we approach, but not quite attain state of the
art performance. The second model can achieve near-perfect accuracy.
| We investigate mapping the hyponymy relation of wordnet to feature vectors |
Recurrent Neural Networks (RNNs) are powerful autoregressive sequence models for learning prevalent patterns in natural language. Yet language generated by RNNs often shows several degenerate characteristics that are uncommon in human language; while fluent, RNN language production can be overly generic, repetitive, and even self-contradictory. We postulate that the objective function optimized by RNN language models, which amounts to the overall perplexity of a text, is not expressive enough to capture the abstract qualities of good generation such as Grice’s Maxims. In this paper, we introduce a general learning framework that can construct a decoding objective better suited for generation. Starting with a generatively trained RNN language model, our framework learns to construct a substantially stronger generator by combining several discriminatively trained models that can collectively address the limitations of RNN generation. Human evaluation demonstrates that text generated by the resulting generator is preferred over that of baselines by a large margin and significantly enhances the overall coherence, style, and information content of the generated text. | We build a stronger natural language generator by discriminatively training scoring functions that rank candidate generations with respect to various qualities of good writing. |
In recent years, the efficiency and even the feasibility of traditional load-balancing policies are challenged by the rapid growth of cloud infrastructure with increasing levels of server heterogeneity and increasing size of cloud services and applications. In such many software-load-balancers heterogeneous systems, traditional solutions, such as JSQ, incur an increasing communication overhead, whereas low-communication alternatives, such as JSQ(d) and the recently proposed JIQ scheme are either unstable or provide poor performance.
We argue that a better low-communication load balancing scheme can be established by allowing each dispatcher to have a different view of the system and keep using JSQ, rather than greedily trying to avoid starvation on a per-decision basis.
accordingly, we introduce the Loosely-Shortest -Queue family of load balancing algorithms. Roughly speaking, in Loosely-shortest -Queue, each dispatcher keeps a different approximation of the server queue lengths and routes jobs to the shortest among them. Communication is used only to update the approximations and make sure that they are not too far from the real queue lengths in expectation. We formally establish the strong stability of any Loosely-Shortest -Queue policy and provide an easy-to-verify sufficient condition for verifying that a policy is Loosely-Shortest -Queue. We further demonstrate that the Loosely-Shortest -Queue approach allows constructing throughput optimal policies with an arbitrarily low communication budget.
Finally, using extensive simulations that consider homogeneous, heterogeneous and highly skewed heterogeneous systems in scenarios with a single dispatcher as well as with multiple dispatchers, we show that the examined Loosely-Shortest -Queue example policies are always stable as dictated by theory. Moreover, it exhibits an appealing performance and significantly outperforms well-known low-communication policies, such as JSQ(d) and JIQ, while using a similar communication budget. | Scalable and low communication load balancing solution for heterogeneous-server multi-dispatcher systems with strong theoretical guarantees and promising empirical results. |
We propose a novel quantitative measure to predict the performance of a deep neural network classifier, where the measure is derived exclusively from the graph structure of the network. We expect that this measure is a fundamental first step in developing a method to evaluate new network architectures and reduce the reliance on the computationally expensive trial and error or "brute force" optimisation processes involved in model selection. The measure is derived in the context of multi-layer perceptrons (MLPs), but the definitions are shown to be useful also in the context of deep convolutional neural networks (CNN), where it is able to estimate and compare the relative performance of different types of neural networks, such as VGG, ResNet, and DenseNet. Our measure is also used to study the effects of some important "hidden" hyper-parameters of the DenseNet architecture, such as number of layers, growth rate and the dimension of 1x1 convolutions in DenseNet-BC. Ultimately, our measure facilitates the optimisation of the DenseNet design, which shows improved results compared to the baseline.
| A quantitative measure to predict the performances of deep neural network models. |
There is a stark disparity between the learning rate schedules used in the practice of large scale machine learning and what are considered admissible learning rate schedules prescribed in the theory of stochastic approximation. Recent results, such as in the 'super-convergence' methods which use oscillating learning rates, serve to emphasize this point even more.
One plausible explanation is that non-convex neural network training procedures are better suited to the use of fundamentally different learning rate schedules, such as the ``cut the learning rate every constant number of epochs'' method (which more closely resembles an exponentially decaying learning rate schedule); note that this widely used schedule is in stark contrast to the polynomial decay schemes prescribed in the stochastic approximation literature, which are indeed shown to be (worst case) optimal for classes of convex optimization problems.
The main contribution of this work shows that the picture is far more nuanced, where we do not even need to move to non-convex optimization to show other learning rate schemes can be far more effective. In fact, even for the simple case of stochastic linear regression with a fixed time horizon, the rate achieved by any polynomial decay scheme is sub-optimal compared to the statistical minimax rate (by a factor of condition number); in contrast the ```''cut the learning rate every constant number of epochs'' provides an exponential improvement (depending only logarithmically on the condition number) compared to any polynomial decay scheme. Finally, it is important to ask if our theoretical insights are somehow fundamentally tied to quadratic loss minimization (where we have circumvented minimax lower bounds for more general convex optimization problems)? Here, we conjecture that recent results which make the gradient norm small at a near optimal rate, for both convex and non-convex optimization, may also provide more insights into learning rate schedules used in practice.
| This paper presents a rigorous study of why practically used learning rate schedules (for a given computational budget) offer significant advantages even though these schemes are not advocated by the classical theory of Stochastic Approximation. |
We present Value Propagation (VProp), a set of parameter-efficient differentiable planning modules built on Value Iteration which can successfully be trained using reinforcement learning to solve unseen tasks, has the capability to generalize to larger map sizes, and can learn to navigate in dynamic environments. We show that the modules enable learning to plan when the environment also includes stochastic elements, providing a cost-efficient learning system to build low-level size-invariant planners for a variety of interactive navigation problems. We evaluate on static and dynamic configurations of MazeBase grid-worlds, with randomly generated environments of several different sizes, and on a StarCraft navigation scenario, with more complex dynamics, and pixels as input. | We present planners based on convnets that are sample-efficient and that generalize to larger instances of navigation and pathfinding problems. |
Learning high-quality word embeddings is of significant importance in achieving better performance in many down-stream learning tasks. On one hand, traditional word embeddings are trained on a large scale corpus for general-purpose tasks, which are often sub-optimal for many domain-specific tasks. On the other hand, many domain-specific tasks do not have a large enough domain corpus to obtain high-quality embeddings. We observe that domains are not isolated and a small domain corpus can leverage the learned knowledge from many past domains to augment that corpus in order to generate high-quality embeddings. In this paper, we formulate the learning of word embeddings as a lifelong learning process. Given knowledge learned from many previous domains and a small new domain corpus, the proposed method can effectively generate new domain embeddings by leveraging a simple but effective algorithm and a meta-learner, where the meta-learner is able to provide word context similarity information at the domain-level. Experimental results demonstrate that the proposed method can effectively learn new domain embeddings from a small corpus and past domain knowledges\footnote{We will release the code after final revisions.}. We also demonstrate that general-purpose embeddings trained from a large scale corpus are sub-optimal in domain-specific tasks. | learning better domain embeddings via lifelong learning and meta-learning |
Parameter pruning is a promising approach for CNN compression and acceleration by eliminating redundant model parameters with tolerable performance loss. Despite its effectiveness, existing regularization-based parameter pruning methods usually drive weights towards zero with large and constant regularization factors, which neglects the fact that the expressiveness of CNNs is fragile and needs a more gentle way of regularization for the networks to adapt during pruning. To solve this problem, we propose a new regularization-based pruning method (named IncReg) to incrementally assign different regularization factors to different weight groups based on their relative importance, whose effectiveness is proved on popular CNNs compared with state-of-the-art methods. | we propose a new regularization-based pruning method (named IncReg) to incrementally assign different regularization factors to different weight groups based on their relative importance. |
Momentum based stochastic gradient methods such as heavy ball (HB) and Nesterov's accelerated gradient descent (NAG) method are widely used in practice for training deep networks and other supervised learning models, as they often provide significant improvements over stochastic gradient descent (SGD). Rigorously speaking, fast gradient methods have provable improvements over gradient descent only for the deterministic case, where the gradients are exact. In the stochastic case, the popular explanations for their wide applicability is that when these fast gradient methods are applied in the stochastic case, they partially mimic their exact gradient counterparts, resulting in some practical gain. This work provides a counterpoint to this belief by proving that there exist simple problem instances where these methods cannot outperform SGD despite the best setting of its parameters. These negative problem instances are, in an informal sense, generic; they do not look like carefully constructed pathological instances. These results suggest (along with empirical evidence) that HB or NAG's practical performance gains are a by-product of minibatching.
Furthermore, this work provides a viable (and provable) alternative, which, on the same set of problem instances, significantly improves over HB, NAG, and SGD's performance. This algorithm, referred to as Accelerated Stochastic Gradient Descent (ASGD), is a simple to implement stochastic algorithm, based on a relatively less popular variant of Nesterov's Acceleration. Extensive empirical results in this paper show that ASGD has performance gains over HB, NAG, and SGD. The code for implementing the ASGD Algorithm can be found at https://github.com/rahulkidambi/AccSGD.
| Existing momentum/acceleration schemes such as heavy ball method and Nesterov's acceleration employed with stochastic gradients do not improve over vanilla stochastic gradient descent, especially when employed with small batch sizes. |
Oversubscription planning (OSP) is the problem of finding plans that maximize the utility value of their end state while staying within a specified cost bound. Recently, it has been shown that OSP problems can be reformulated as classical planning problems with multiple cost functions but no utilities. Here we take advantage of this reformulation to show that OSP problems can be solved optimally using the A* search algorithm, in contrast to previous approaches that have used variations on branch-and-bound search. This allows many powerful techniques developed for classical planning to be applied to OSP problems. We also introduce novel bound-sensitive heuristics, which are able to reason about the primary cost of a solution while taking into account secondary cost functions and bounds, to provide superior guidance compared to heuristics that do not take these bounds into account. We implement two such bound-sensitive variants of existing classical planning heuristics, and show experimentally that the resulting search is significantly more informed than comparable heuristics that do not consider bounds. | We show that oversubscription planning tasks can be solved using A* and introduce novel bound-sensitive heuristics for oversubscription planning tasks. |
Previous work on adversarially robust neural networks requires large training sets and computationally expensive training procedures. On the other hand, few-shot learning methods are highly vulnerable to adversarial examples. The goal of our work is to produce networks which both perform well at few-shot tasks and are simultaneously robust to adversarial examples. We adapt adversarial training for meta-learning, we adapt robust architectural features to small networks for meta-learning, we test pre-processing defenses as an alternative to adversarial training for meta-learning, and we investigate the advantages of robust meta-learning over robust transfer-learning for few-shot tasks. This work provides a thorough analysis of adversarially robust methods in the context of meta-learning, and we lay the foundation for future work on defenses for few-shot tasks. | We develop meta-learning methods for adversarially robust few-shot learning. |
Many of our core assumptions about how neural networks operate remain empirically untested. One common assumption is that convolutional neural networks need to be stable to small translations and deformations to solve image recognition tasks. For many years, this stability was baked into CNN architectures by incorporating interleaved pooling layers. Recently, however, interleaved pooling has largely been abandoned. This raises a number of questions: Are our intuitions about deformation stability right at all? Is it important? Is pooling necessary for deformation invariance? If not, how is deformation invariance achieved in its absence? In this work, we rigorously test these questions, and find that deformation stability in convolutional networks is more nuanced than it first appears: (1) Deformation invariance is not a binary property, but rather that different tasks require different degrees of deformation stability at different layers. (2) Deformation stability is not a fixed property of a network and is heavily adjusted over the course of training, largely through the smoothness of the convolutional filters. (3) Interleaved pooling layers are neither necessary nor sufficient for achieving the optimal form of deformation stability for natural image classification. (4) Pooling confers \emph{too much} deformation stability for image classification at initialization, and during training, networks have to learn to \emph{counteract} this inductive bias. Together, these findings provide new insights into the role of interleaved pooling and deformation invariance in CNNs, and demonstrate the importance of rigorous empirical testing of even our most basic assumptions about the working of neural networks. | We find that pooling alone does not determine deformation stability in CNNs and that filter smoothness plays an important role in determining stability. |
Deep neural networks (DNNs) have been shown to over-fit a dataset when being trained with noisy labels for a long enough time. To overcome this problem, we present a simple and effective method self-ensemble label filtering (SELF) to progressively filter out the wrong labels during training. Our method improves the task performance by gradually allowing supervision only from the potentially non-noisy (clean) labels and stops learning on the filtered noisy labels. For the filtering, we form running averages of predictions over the entire training dataset using the network output at different training epochs. We show that these ensemble estimates yield more accurate identification of inconsistent predictions throughout training than the single estimates of the network at the most recent training epoch. While filtered samples are removed entirely from the supervised training loss, we dynamically leverage them via semi-supervised learning in the unsupervised loss. We demonstrate the positive effect of such an approach on various image classification tasks under both symmetric and asymmetric label noise and at different noise ratios. It substantially outperforms all previous works on noise-aware learning across different datasets and can be applied to a broad set of network architectures. | We propose a self-ensemble framework to train more robust deep learning models under noisy labeled datasets. |
Long training times of deep neural networks are a bottleneck in machine learning research. The major impediment to fast training is the quadratic growth of both memory and compute requirements of dense and convolutional layers with respect to their information bandwidth. Recently, training `a priori' sparse networks has been proposed as a method for allowing layers to retain high information bandwidth, while keeping memory and compute low. However, the choice of which sparse topology should be used in these networks is unclear. In this work, we provide a theoretical foundation for the choice of intra-layer topology. First, we derive a new sparse neural network initialization scheme that allows us to explore the space of very deep sparse networks. Next, we evaluate several topologies and show that seemingly similar topologies can often have a large difference in attainable accuracy. To explain these differences, we develop a data-free heuristic that can evaluate a topology independently from the dataset the network will be trained on. We then derive a set of requirements that make a good topology, and arrive at a single topology that satisfies all of them. | We investigate pruning DNNs before training and provide an answer to which topology should be used for training a priori sparse networks. |
Deep learning models require extensive architecture design exploration and hyperparameter optimization to perform well on a given task. The exploration of the model design space is often made by a human expert, and optimized using a combination of grid search and search heuristics over a large space of possible choices. Neural Architecture Search (NAS) is a Reinforcement Learning approach that has been proposed to automate architecture design. NAS has been successfully applied to generate Neural Networks that rival the best human-designed architectures. However, NAS requires sampling, constructing, and training hundreds to thousands of models to achieve well-performing architectures. This procedure needs to be executed from scratch for each new task. The application of NAS to a wide set of tasks currently lacks a way to transfer generalizable knowledge across tasks.
In this paper, we present the Multitask Neural Model Search (MNMS) controller. Our goal is to learn a generalizable framework that can condition model construction on successful model searches for previously seen tasks, thus significantly speeding up the search for new tasks. We demonstrate that MNMS can conduct an automated architecture search for multiple tasks simultaneously while still learning well-performing, specialized models for each task. We then show that pre-trained MNMS controllers can transfer learning to new tasks. By leveraging knowledge from previous searches, we find that pre-trained MNMS models start from a better location in the search space and reduce search time on unseen tasks, while still discovering models that outperform published human-designed models. | We present Multitask Neural Model Search, a Meta-learner that can design models for multiple tasks simultaneously and transfer learning to unseen tasks. |
This work studies the problem of modeling non-linear visual processes by leveraging deep generative architectures for learning linear, Gaussian models of observed sequences. We propose a joint learning framework, combining a multivariate autoregressive model and deep convolutional generative networks. After justification of theoretical assumptions of inearization, we propose an architecture that allows Variational Autoencoders and Generative Adversarial Networks to simultaneously learn the non-linear observation as well as the linear state-transition model from a sequence of observed frames. Finally, we demonstrate our approach on conceptual toy examples and dynamic textures. | We model non-linear visual processes as autoregressive noise via generative deep learning. |
Partial differential equations (PDEs) play a prominent role in many disciplines such as applied mathematics, physics, chemistry, material science, computer science, etc. PDEs are commonly derived based on physical laws or empirical observations. However, the governing equations for many complex systems in modern applications are still not fully known. With the rapid development of sensors, computational power, and data storage in the past decade, huge quantities of data can be easily collected and efficiently stored. Such vast quantity of data offers new opportunities for data-driven discovery of hidden physical laws. Inspired by the latest development of neural network designs in deep learning, we propose a new feed-forward deep network, called PDE-Net, to fulfill two objectives at the same time: to accurately predict dynamics of complex systems and to uncover the underlying hidden PDE models. The basic idea of the proposed PDE-Net is to learn differential operators by learning convolution kernels (filters), and apply neural networks or other machine learning methods to approximate the unknown nonlinear responses. Comparing with existing approaches, which either assume the form of the nonlinear response is known or fix certain finite difference approximations of differential operators, our approach has the most flexibility by learning both differential operators and the nonlinear responses. A special feature of the proposed PDE-Net is that all filters are properly constrained, which enables us to easily identify the governing PDE models while still maintaining the expressive and predictive power of the network. These constrains are carefully designed by fully exploiting the relation between the orders of differential operators and the orders of sum rules of filters (an important concept originated from wavelet theory). We also discuss relations of the PDE-Net with some existing networks in computer vision such as Network-In-Network (NIN) and Residual Neural Network (ResNet). Numerical experiments show that the PDE-Net has the potential to uncover the hidden PDE of the observed dynamics, and predict the dynamical behavior for a relatively long time, even in a noisy environment. | This paper proposes a new feed-forward network, call PDE-Net, to learn PDEs from data. |
Each training step for a variational autoencoder (VAE) requires us to sample from the approximate posterior, so we usually choose simple (e.g. factorised) approximate posteriors in which sampling is an efficient computation that fully exploits GPU parallelism. However, such simple approximate posteriors are often insufficient, as they eliminate statistical dependencies in the posterior. While it is possible to use normalizing flow approximate posteriors for continuous latents, there is nothing analogous for discrete latents. The most natural approach to model discrete dependencies is an autoregressive distribution, but sampling from such distributions is inherently sequential and thus slow. We develop a fast, parallel sampling procedure for autoregressive distributions based on fixed-point iterations which enables efficient and accurate variational inference in discrete state-space models. To optimize the variational bound, we considered two ways to evaluate probabilities: inserting the relaxed samples directly into the pmf for the discrete distribution, or converting to continuous logistic latent variables and interpreting the K-step fixed-point iterations as a normalizing flow. We found that converting to continuous latent variables gave considerable additional scope for mismatch between the true and approximate posteriors, which resulted in biased inferences, we thus used the former approach. We tested our approach on the neuroscience problem of inferring discrete spiking activity from noisy calcium-imaging data, and found that it gave accurate connectivity estimates in an order of magnitude less time. | We give a fast normalising-flow like sampling procedure for discrete latent variable models. |
Deep neural networks (DNNs) had great success on NLP tasks such as language modeling, machine translation and certain question answering (QA) tasks. However, the success is limited at more knowledge intensive tasks such as QA from a big corpus. Existing end-to-end deep QA models (Miller et al., 2016; Weston et al., 2014) need to read the entire text after observing the question, and therefore their complexity in responding a question is linear in the text size. This is prohibitive for practical tasks such as QA from Wikipedia, a novel, or the Web. We propose to solve this scalability issue by using symbolic meaning representations, which can be indexed and retrieved efficiently with complexity that is independent of the text size. More specifically, we use sequence-to-sequence models to encode knowledge symbolically and generate programs to answer questions from the encoded knowledge. We apply our approach, called the N-Gram Machine (NGM), to the bAbI tasks (Weston et al., 2015) and a special version of them (“life-long bAbI”) which has stories of up to 10 million sentences. Our experiments show that NGM can successfully solve both of these tasks accurately and efficiently. Unlike fully differentiable memory models, NGM’s time complexity and answering quality are not affected by the story length. The whole system of NGM is trained end-to-end with REINFORCE (Williams, 1992). To avoid high variance in gradient estimation, which is typical in discrete latent variable models, we use beam search instead of sampling. To tackle the exponentially large search space, we use a stabilized auto-encoding objective and a structure tweak procedure to iteratively reduce and refine the search space.
| We propose a framework that learns to encode knowledge symbolically and generate programs to reason about the encoded knowledge. |
We propose to use a meta-learning objective that maximizes the speed of transfer on a modified distribution to learn how to modularize acquired knowledge. In particular, we focus on how to factor a joint distribution into appropriate conditionals, consistent with the causal directions. We explain when this can work, using the assumption that the changes in distributions are localized (e.g. to one of the marginals, for example due to an intervention on one of the variables). We prove that under this assumption of localized changes in causal mechanisms, the correct causal graph will tend to have only a few of its parameters with non-zero gradient, i.e. that need to be adapted (those of the modified variables). We argue and observe experimentally that this leads to faster adaptation, and use this property to define a meta-learning surrogate score which, in addition to a continuous parametrization of graphs, would favour correct causal graphs. Finally, motivated by the AI agent point of view (e.g. of a robot discovering its environment autonomously), we consider how the same objective can discover the causal variables themselves, as a transformation of observed low-level variables with no causal meaning. Experiments in the two-variable case validate the proposed ideas and theoretical results. | This paper proposes a meta-learning objective based on speed of adaptation to transfer distributions to discover a modular decomposition and causal variables. |
Continual learning is a longstanding goal of artificial intelligence, but is often counfounded by catastrophic forgetting that prevents neural networks from learning tasks sequentially. Previous methods in continual learning have demonstrated how to mitigate catastrophic forgetting, and learn new tasks while retaining performance on the previous tasks. We analyze catastrophic forgetting from the perspective of change in classifier likelihood and propose a simple L1 minimization criterion which can be adapted to different use cases. We further investigate two ways to minimize forgetting as quantified by this criterion and propose strategies to achieve finer control over forgetting. Finally, we evaluate our strategies on 3 datasets of varying difficulty and demonstrate improvements over previously known L2 strategies for mitigating catastrophic forgetting. | Another perspective on catastrophic forgetting |
We propose an approach to construct realistic 3D facial morphable models (3DMM) that allows an intuitive facial attribute
editing workflow. Current face modeling methods using 3DMM suffer from the lack of local control. We thus create a 3DMM by
combining local part-based 3DMM for the eyes, nose, mouth, ears, and facial mask regions. Our local PCA-based approach
uses a novel method to select the best eigenvectors from the local 3DMM to ensure that the combined 3DMM is expressive
while allowing accurate reconstruction. The editing controls we provide to the user are intuitive as they are extracted from
anthropometric measurements found in the literature. Out of a large set of possible anthropometric measurements, we filter the
ones that have meaningful generative power given the face data set. We bind the measurements to the part-based 3DMM through
mapping matrices derived from our data set of facial scans. Our part-based 3DMM is compact yet accurate, and compared to
other 3DMM methods, it provides a new trade-off between local and global control. We tested our approach on a data set of 135
scans used to derive the 3DMM, plus 19 scans that served for validation. The results show that our part-based 3DMM approach
has excellent generative properties and allows intuitive local control to the user.
| We propose an approach to construct realistic 3D facial morphable models (3DMM) that allows an intuitive facial attribute editing workflow by selecting the best sets of eigenvectors and anthropometric measurements. |
We review eight machine learning classification algorithms to analyze Electroencephalographic (EEG) signals in order to distinguish EEG patterns associated with five basic educational tasks. There is a large variety of classifiers being used in this EEG-based Brain-Computer Interface (BCI) field. While previous EEG experiments used several classifiers in the same experiments or reviewed different algorithms on datasets from different experiments, our approach focuses on review eight classifier categories on the same dataset, including linear classifiers, non-linear Bayesian classifiers, nearest neighbour classifiers, ensemble methods, adaptive classifiers, tensor classifiers, transfer learning and deep learning. Besides, we intend to find an approach which can run smoothly on the current mainstream personal computers and smartphones. The empirical evaluation demonstrated that Random Forest and LSTM (Long Short-Term Memory) outperform other approaches. We used a data set which users were conducting five frequently-conduct learning-related tasks, including reading, writing, and typing. Results showed that these best two algorithms could correctly classify different users with an accuracy increase of 5% to 9%, use each task independently. Within each subject, the tasks could be recognized with an accuracy increase of 4% to 7%, compared with other approaches. This work suggests that Random Forest could be a recommended approach (fast and accurate) for current mainstream hardware, while LSTM has the potential to be the first-choice approach when the mainstream computers and smartphones can process more data in a shorter time. | Two Algorithms outperformed eight others on a EEG-based BCI experiment |
Multi-agent reinforcement learning offers a way to study how communication could emerge in communities of agents needing to solve specific problems. In this paper, we study the emergence of communication in the negotiation environment, a semi-cooperative model of agent interaction. We introduce two communication protocols - one grounded in the semantics of the game, and one which is a priori ungrounded. We show that self-interested agents can use the pre-grounded communication channel to negotiate fairly, but are unable to effectively use the ungrounded, cheap talk channel to do the same. However, prosocial agents do learn to use cheap talk to find an optimal negotiating strategy, suggesting that cooperation is necessary for language to emerge. We also study communication behaviour in a setting where one agent interacts with agents in a community with different levels of prosociality and show how agent identifiability can aid negotiation. | We teach agents to negotiate using only reinforcement learning; selfish agents can do so, but only using a trustworthy communication channel, and prosocial agents can negotiate using cheap talk. |
The goal of few-shot learning is to learn a classifier that generalizes well even when trained with a limited number of training instances per class. The recently introduced meta-learning approaches tackle this problem by learning a generic classifier across a large number of multiclass classification tasks and generalizing the model to a new task. Yet, even with such meta-learning, the low-data problem in the novel classification task still remains. In this paper, we propose Transductive Propagation Network (TPN), a novel meta-learning framework for transductive inference that classifies the entire test set at once to alleviate the low-data problem. Specifically, we propose to learn to propagate labels from labeled instances to unlabeled test instances, by learning a graph construction module that exploits the manifold structure in the data. TPN jointly learns both the parameters of feature embedding and the graph construction in an end-to-end manner. We validate TPN on multiple benchmark datasets, on which it largely outperforms existing few-shot learning approaches and achieves the state-of-the-art results. | We propose a novel meta-learning framework for transductive inference that classifies the entire test set at once to alleviate the low-data problem. |
We describe the use of an automated scheduling system for observation policy design and to schedule operations of the NASA (National Aeronautics and Space Administration) ECOSystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS). We describe the adaptation of the Compressed Large-scale Activity Scheduler and Planner (CLASP) scheduling system to the ECOSTRESS scheduling problem, highlighting multiple use cases for automated scheduling and several challenges for the scheduling technology: handling long-term campaigns with changing information, Mass Storage Unit Ring Buffer operations challenges, and orbit uncertainty. The described scheduling system has been used for operations of the ECOSTRESS instrument since its nominal operations start July 2018 and is expected to operate until mission end in Summer 2019. | We describe the use of an automated scheduling system for observation policy design and to schedule operations of NASA's ECOSTRESS mission. |
Adversarial examples are modified samples that preserve original image structures but deviate classifiers. Researchers have put efforts into developing methods for generating adversarial examples and finding out origins. Past research put much attention on decision boundary changes caused by these methods. This paper, in contrast, discusses the origin of adversarial examples from a more underlying knowledge representation point of view. Human beings can learn and classify prototypes as well as transformations of objects. While neural networks store learned knowledge in a more hybrid way of combining all prototypes and transformations as a whole distribution. Hybrid storage may lead to lower distances between different classes so that small modifications can mislead the classifier. A one-step distribution imitation method is designed to imitate distribution of the nearest different class neighbor. Experiments show that simply by imitating distributions from a training set without any knowledge of the classifier can still lead to obvious impacts on classification results from deep networks. It also implies that adversarial examples can be in more forms than small perturbations. Potential ways of alleviating adversarial examples are discussed from the representation point of view. The first path is to change the encoding of data sent to the training step. Training data that are more prototypical can help seize more robust and accurate structural knowledge. The second path requires constructing learning frameworks with improved representations. | Hybird storage and representation of learned knowledge may be a reason for adversarial examples. |
Differently from the popular Deep Q-Network (DQN) learning, Alternating Q-learning (AltQ) does not fully fit a target Q-function at each iteration, and is generally known to be unstable and inefficient. Limited applications of AltQ mostly rely on substantially altering the algorithm architecture in order to improve its performance. Although Adam appears to be a natural solution, its performance in AltQ has rarely been studied before. In this paper, we first provide a solid exploration on how well AltQ performs with Adam. We then take a further step to improve the implementation by adopting the technique of parameter restart. More specifically, the proposed algorithms are tested on a batch of Atari 2600 games and exhibit superior performance than the DQN learning method. The convergence rate of the slightly modified version of the proposed algorithms is characterized under the linear function approximation. To the best of our knowledge, this is the first theoretical study on the Adam-type algorithms in Q-learning. | New Experiments and Theory for Adam Based Q-Learning |
In search for more accurate predictive models, we customize capsule networks for the learning to diagnose problem. We also propose Spectral Capsule Networks, a novel variation of capsule networks, that converge faster than capsule network with EM routing. Spectral capsule networks consist of spatial coincidence filters that detect entities based on the alignment of extracted features on a one-dimensional linear subspace. Experiments on a public benchmark learning to diagnose dataset not only shows the success of capsule networks on this task, but also confirm the faster convergence of the spectral capsule networks. | A new capsule network that converges faster on our healthcare benchmark experiments. |
One of the big challenges in machine learning applications is that training data can be different from the real-world data faced by the algorithm. In language modeling, users’ language (e.g. in private messaging) could change in a year and be completely different from what we observe in publicly available data. At the same time, public data can be used for obtaining general knowledge (i.e. general model of English). We study approaches to distributed fine-tuning of a general model on user private data with the additional requirements of maintaining the quality on the general data and minimization of communication costs. We propose a novel technique that significantly improves prediction quality on users’ language compared to a general model and outperforms gradient compression methods in terms of communication efficiency. The proposed procedure is fast and leads to an almost 70% perplexity reduction and 8.7 percentage point improvement in keystroke saving rate on informal English texts. Finally, we propose an experimental framework for evaluating differential privacy of distributed training of language models and show that our approach has good privacy guarantees. | We propose a method of distributed fine-tuning of language models on user devices without collection of private data |
We propose that approximate Bayesian algorithms should optimize a new criterion, directly derived from the loss, to calculate their approximate posterior which we refer to as pseudo-posterior. Unlike standard variational inference which optimizes a lower bound on the log marginal likelihood, the new algorithms can be analyzed to provide loss guarantees on the predictions with the pseudo-posterior. Our criterion can be used to derive new sparse Gaussian process algorithms that have error guarantees applicable to various likelihoods. | This paper utilizes the analysis of Lipschitz loss on a bounded hypothesis space to derive new ERM-type algorithms with strong performance guarantees that can be applied to the non-conjugate sparse GP model. |
In this paper, we propose a novel regularization method, RotationOut, for neural networks.
Different from Dropout that handles each neuron/channel independently, RotationOut regards its input layer as an entire vector and introduces regularization by randomly rotating the vector.
RotationOut can also be used in convolutional layers and recurrent layers with a small modification.
We further use a noise analysis method to interpret the difference between RotationOut and Dropout in co-adaptation reduction.
Using this method, we also show how to use RotationOut/Dropout together with Batch Normalization.
Extensive experiments in vision and language tasks are conducted to show the effectiveness of the proposed method.
Codes will be available. | We propose a regularization method for neural network and a noise analysis method |
Formulating the reinforcement learning (RL) problem in the framework of probabilistic inference not only offers a new perspective about RL, but also yields practical algorithms that are more robust and easier to train. While this connection between RL and probabilistic inference has been extensively studied in the single-agent setting, it has not yet been fully understood in the multi-agent setup. In this paper, we pose the problem of multi-agent reinforcement learning as the problem of performing inference in a particular graphical model. We model the environment, as seen by each of the agents, using separate but related Markov decision processes. We derive a practical off-policy maximum-entropy actor-critic algorithm that we call Multi-agent Soft Actor-Critic (MA-SAC) for performing approximate inference in the proposed model using variational inference. MA-SAC can be employed in both cooperative and competitive settings. Through experiments, we demonstrate that MA-SAC outperforms a strong baseline on several multi-agent scenarios. While MA-SAC is one resultant multi-agent RL algorithm that can be derived from the proposed probabilistic framework, our work provides a unified view of maximum-entropy algorithms in the multi-agent setting. | A probabilistic framework for multi-agent reinforcement learning |
Sorting input objects is an important step in many machine learning pipelines. However, the sorting operator is non-differentiable with respect to its inputs, which prohibits end-to-end gradient-based optimization. In this work, we propose NeuralSort, a general-purpose continuous relaxation of the output of the sorting operator from permutation matrices to the set of unimodal row-stochastic matrices, where every row sums to one and has a distinct argmax. This relaxation permits straight-through optimization of any computational graph involve a sorting operation. Further, we use this relaxation to enable gradient-based stochastic optimization over the combinatorially large space of permutations by deriving a reparameterized gradient estimator for the Plackett-Luce family of distributions over permutations. We demonstrate the usefulness of our framework on three tasks that require learning semantic orderings of high-dimensional objects, including a fully differentiable, parameterized extension of the k-nearest neighbors algorithm | We provide a continuous relaxation to the sorting operator, enabling end-to-end, gradient-based stochastic optimization. |
Transferring knowledge across tasks to improve data-efficiency is one of
the open key challenges in the area of global optimization algorithms. Readily
available algorithms are typically designed to be universal optimizers and, thus,
often suboptimal for specific tasks. We propose a novel transfer learning method to
obtain customized optimizers within the well-established framework of Bayesian
optimization, allowing our algorithm to utilize the proven generalization
capabilities of Gaussian processes. Using reinforcement learning to meta-train an
acquisition function (AF) on a set of related tasks, the proposed method learns to
extract implicit structural information and to exploit it for improved data-efficiency.
We present experiments on a sim-to-real transfer task as well as on several simulated
functions and two hyperparameter search problems. The results show that our
algorithm (1) automatically identifies structural properties of objective functions
from available source tasks or simulations, (2) performs favourably in settings with
both scarse and abundant source data, and (3) falls back to the performance level
of general AFs if no structure is present. | We perform efficient and flexible transfer learning in the framework of Bayesian optimization through meta-learned neural acquisition functions. |
We study the evolution of internal representations during deep neural network (DNN) training, aiming to demystify the compression aspect of the information bottleneck theory. The theory suggests that DNN training comprises a rapid fitting phase followed by a slower compression phase, in which the mutual information I(X;T) between the input X and internal representations T decreases. Several papers observe compression of estimated mutual information on different DNN models, but the true I(X;T) over these networks is provably either constant (discrete X) or infinite (continuous X). This work explains the discrepancy between theory and experiments, and clarifies what was actually measured by these past works. To this end, we introduce an auxiliary (noisy) DNN framework for which I(X;T) is a meaningful quantity that depends on the network's parameters. This noisy framework is shown to be a good proxy for the original (deterministic) DNN both in terms of performance and the learned representations. We then develop a rigorous estimator for I(X;T) in noisy DNNs and observe compression in various models. By relating I(X;T) in the noisy DNN to an information-theoretic communication problem, we show that compression is driven by the progressive clustering of hidden representations of inputs from the same class. Several methods to directly monitor clustering of hidden representations, both in noisy and deterministic DNNs, are used to show that meaningful clusters form in the T space. Finally, we return to the estimator of I(X;T) employed in past works, and demonstrate that while it fails to capture the true (vacuous) mutual information, it does serve as a measure for clustering. This clarifies the past observations of compression and isolates the geometric clustering of hidden representations as the true phenomenon of interest. | Deterministic deep neural networks do not discard information, but they do cluster their inputs. |
A central challenge in multi-agent reinforcement learning is the induction of coordination between agents of a team. In this work, we investigate how to promote inter-agent coordination using policy regularization and discuss two possible avenues respectively based on inter-agent modelling and synchronized sub-policy selection. We test each approach in four challenging continuous control tasks with sparse rewards and compare them against three baselines including MADDPG, a state-of-the-art multi-agent reinforcement learning algorithm. To ensure a fair comparison, we rely on a thorough hyper-parameter selection and training methodology that allows a fixed hyper-parameter search budget for each algorithm and environment. We consequently assess both the hyper-parameter sensitivity, sample-efficiency and asymptotic performance of each learning method. Our experiments show that the proposed methods lead to significant improvements on cooperative problems. We further analyse the effects of the proposed regularizations on the behaviors learned by the agents. | We propose regularization objectives for multi-agent RL algorithms that foster coordination on cooperative tasks. |
Multimodal sentiment analysis is a core research area that studies speaker sentiment expressed from the language, visual, and acoustic modalities. The central challenge in multimodal learning involves inferring joint representations that can process and relate information from these modalities. However, existing work learns joint representations using multiple modalities as input and may be sensitive to noisy or missing modalities at test time. With the recent success of sequence to sequence models in machine translation, there is an opportunity to explore new ways of learning joint representations that may not require all input modalities at test time. In this paper, we propose a method to learn robust joint representations by translating between modalities. Our method is based on the key insight that translation from a source to a target modality provides a method of learning joint representations using only the source modality as input. We augment modality translations with a cycle consistency loss to ensure that our joint representations retain maximal information from all modalities. Once our translation model is trained with paired multimodal data, we only need data from the source modality at test-time for prediction. This ensures that our model remains robust from perturbations or missing target modalities. We train our model with a coupled translation-prediction objective and it achieves new state-of-the-art results on multimodal sentiment analysis datasets: CMU-MOSI, ICT-MMMO, and YouTube. Additional experiments show that our model learns increasingly discriminative joint representations with more input modalities while maintaining robustness to perturbations of all other modalities. | We present a model that learns robust joint representations by performing hierarchical cyclic translations between multiple modalities. |
The geometric properties of loss surfaces, such as the local flatness of a solution, are associated with generalization in deep learning. The Hessian is often used to understand these geometric properties. We investigate the differences between the eigenvalues of the neural network Hessian evaluated over the empirical dataset, the Empirical Hessian, and the eigenvalues of the Hessian under the data generating distribution, which we term the True Hessian. Under mild assumptions, we use random matrix theory to show that the True Hessian has eigenvalues of smaller absolute value than the Empirical Hessian. We support these results for different SGD schedules on both a 110-Layer ResNet and VGG-16. To perform these experiments we propose a framework for spectral visualization, based on GPU accelerated stochastic Lanczos quadrature. This approach is an order of magnitude faster than state-of-the-art methods for spectral visualization, and can be generically used to investigate the spectral properties of matrices in deep learning. | Understanding the neural network Hessian eigenvalues under the data generating distribution. |
Summarization of long sequences into a concise statement is a core problem in natural language processing, requiring non-trivial understanding of the input. Based on the promising results of graph neural networks on highly structured data, we develop a framework to extend existing sequence encoders with a graph component that can reason about long-distance relationships in weakly structured data such as text. In an extensive evaluation, we show that the resulting hybrid sequence-graph models outperform both pure sequence models as well as pure graph models on a range of summarization tasks. | One simple trick to improve sequence models: Compose them with a graph model |
In probabilistic classification, a discriminative model based on Gaussian mixture exhibits flexible fitting capability. Nevertheless, it is difficult to determine the number of components. We propose a sparse classifier based on a discriminative Gaussian mixture model (GMM), which is named sparse discriminative Gaussian mixture (SDGM). In the SDGM, a GMM-based discriminative model is trained by sparse Bayesian learning. This learning algorithm improves the generalization capability by obtaining a sparse solution and automatically determines the number of components by removing redundant components. The SDGM can be embedded into neural networks (NNs) such as convolutional NNs and can be trained in an end-to-end manner. Experimental results indicated that the proposed method prevented overfitting by obtaining sparsity. Furthermore, we demonstrated that the proposed method outperformed a fully connected layer with the softmax function in certain cases when it was used as the last layer of a deep NN. | A sparse classifier based on a discriminative Gaussian mixture model, which can also be embedded into a neural network. |
We recently observed that convolutional filters initialized
farthest apart from each other using offthe-
shelf pre-computed Grassmannian subspace
packing codebooks performed surprisingly well
across many datasets. Through this short paper,
we’d like to disseminate some initial results in this
regard in the hope that we stimulate the curiosity
of the deep-learning community towards considering
classical Grassmannian subspace packing
results as a source of new ideas for more efficient
initialization strategies. | Initialize weights using off-the-shelf Grassmannian codebooks, get faster training and better accuracy |