abstract
stringlengths
42
2.09k
We define the random magnetic Laplacien with spatial white noise as magnetic field on the two-dimensional torus using paracontrolled calculus. It yields a random self-adjoint operator with pure point spectrum and domain a random subspace of nonsmooth functions in L 2. We give sharp bounds on the eigenvalues which imply an almost sure Weyl-type law.
Searches for gravitational-wave counterparts have been going in earnest since GW170817 and the discovery of AT2017gfo. Since then, the lack of detection of other optical counterparts connected to binary neutron star or black hole - neutron star candidates has highlighted the need for a better discrimination criterion to support this effort. At the moment, the low-latency gravitational-wave alerts contain preliminary information about the binary properties and, hence, on whether a detected binary might have an electromagnetic counterpart. The current alert method is a classifier that estimates the probability that there is a debris disc outside the black hole created during the merger as well as the probability of a signal being a binary neutron star, a black hole - neutron star, a binary black hole or of terrestrial origin. In this work, we expand upon this approach to predict both the ejecta properties and provide contours of potential lightcurves for these events in order to improve follow-up observation strategy. The various sources of uncertainty are discussed, and we conclude that our ignorance about the ejecta composition and the insufficient constraint of the binary parameters, by the low-latency pipelines, represent the main limitations. To validate the method, we test our approach on real events from the second and third Advanced LIGO-Virgo observing runs.
Fully implicit Runge-Kutta (IRK) methods have many desirable properties as time integration schemes in terms of accuracy and stability, but high-order IRK methods are not commonly used in practice with numerical PDEs due to the difficulty of solving the stage equations. This paper introduces a theoretical and algorithmic preconditioning framework for solving the systems of equations that arise from IRK methods applied to linear numerical PDEs (without algebraic constraints). This framework also naturally applies to discontinuous Galerkin discretizations in time. Under quite general assumptions on the spatial discretization that yield stable time integration, the preconditioned operator is proven to have condition number bounded by a small, order-one constant, independent of the spatial mesh and time-step size, and with only weak dependence on number of stages/polynomial order; for example, the preconditioned operator for 10th-order Gauss IRK has condition number less than two, independent of the spatial discretization and time step. The new method can be used with arbitrary existing preconditioners for backward Euler-type time stepping schemes, and is amenable to the use of three-term recursion Krylov methods when the underlying spatial discretization is symmetric. The new method is demonstrated to be effective on various high-order finite-difference and finite-element discretizations of linear parabolic and hyperbolic problems, demonstrating fast, scalable solution of up to 10th order accuracy. The new method consistently outperforms existing block preconditioning approaches, and in several cases, the new method can achieve 4th-order accuracy using Gauss integration with roughly half the number of preconditioner applications and wallclock time as required using standard diagonally implicit RK methods.
To successfully negotiate a deal, it is not enough to communicate fluently: pragmatic planning of persuasive negotiation strategies is essential. While modern dialogue agents excel at generating fluent sentences, they still lack pragmatic grounding and cannot reason strategically. We present DialoGraph, a negotiation system that incorporates pragmatic strategies in a negotiation dialogue using graph neural networks. DialoGraph explicitly incorporates dependencies between sequences of strategies to enable improved and interpretable prediction of next optimal strategies, given the dialogue context. Our graph-based method outperforms prior state-of-the-art negotiation models both in the accuracy of strategy/dialogue act prediction and in the quality of downstream dialogue response generation. We qualitatively show further benefits of learned strategy-graphs in providing explicit associations between effective negotiation strategies over the course of the dialogue, leading to interpretable and strategic dialogues.
We propose a systematic analysis method for identifying essential parameters in various linear and nonlinear response tensors without which they vanish. By using the Keldysh formalism and the Chebyshev polynomial expansion method, the response tensors are decomposed into the model-independent and dependent parts, in which the latter is utilized to extract the essential parameters. An application of the method is demonstrated by analyzing the nonlinear Hall effect in the ferroelectric SnTe monolayer for example. It is shown that in this example the second-neighbor hopping is essential for the nonlinear Hall effect whereas the spin-orbit coupling is unnecessary. Moreover, by analyzing terms contributing to the essential parameters in the lowest order, the appearance of the nonlinear Hall effect can be interpreted by the subsequent two processes: the orbital magneto-current effect and the linear anomalous Hall effect by the induced orbital magnetization. In this way, the present method provides a microscopic picture of responses. By combining with computational analysis, it stimulates further discoveries of anomalous responses by filling in a missing link among hidden degrees of freedom in a wide variety of materials.
It is a long-standing objective to ease the computation burden incurred by the decision making process. Identification of this mechanism's sensitivity to simplification has tremendous ramifications. Yet, algorithms for decision making under uncertainty usually lean on approximations or heuristics without quantifying their effect. Therefore, challenging scenarios could severely impair the performance of such methods. In this paper, we extend the decision making mechanism to the whole by removing standard approximations and considering all previously suppressed stochastic sources of variability. On top of this extension, our key contribution is a novel framework to simplify decision making while assessing and controlling online the simplification's impact. Furthermore, we present novel stochastic bounds on the return and characterize online the effect of simplification using this framework on a particular simplification technique - reducing the number of samples in belief representation for planning. Finally, we verify the advantages of our approach through extensive simulations.
For the development of successful share trading strategies, forecasting the course of action of the stock market index is important. Effective prediction of closing stock prices could guarantee investors attractive benefits. Machine learning algorithms have the ability to process and forecast almost reliable closing prices for historical stock patterns. In this article, we intensively studied NASDAQ stock market and targeted to choose the portfolio of ten different companies belongs to different sectors. The objective is to compute opening price of next day stock using historical data. To fulfill this task nine different Machine Learning regressor applied on this data and evaluated using MSE and R2 as performance metric.
Ensuring the privacy of research participants is vital, even more so in healthcare environments. Deep learning approaches to neuroimaging require large datasets, and this often necessitates sharing data between multiple sites, which is antithetical to the privacy objectives. Federated learning is a commonly proposed solution to this problem. It circumvents the need for data sharing by sharing parameters during the training process. However, we demonstrate that allowing access to parameters may leak private information even if data is never directly shared. In particular, we show that it is possible to infer if a sample was used to train the model given only access to the model prediction (black-box) or access to the model itself (white-box) and some leaked samples from the training data distribution. Such attacks are commonly referred to as Membership Inference attacks. We show realistic Membership Inference attacks on deep learning models trained for 3D neuroimaging tasks in a centralized as well as decentralized setup. We demonstrate feasible attacks on brain age prediction models (deep learning models that predict a person's age from their brain MRI scan). We correctly identified whether an MRI scan was used in model training with a 60% to over 80% success rate depending on model complexity and security assumptions.
An intrinsic antiferromagnetic topological insulator $\mathrm{MnBi_2Te_4}$ can be realized by intercalating Mn-Te bilayer chain in a topological insulator, $\mathrm{Bi_2Te_3}$. $\mathrm{MnBi_2Te_4}$ provides not only a stable platform to demonstrate exotic physical phenomena, but also easy tunability of the physical properties. For example, inserting more $\mathrm{Bi_2Te_3}$ layers in between two adjacent $\mathrm{MnBi_2Te_4}$ weakens the interlayer magnetic interactions between the $\mathrm{MnBi_2Te_4}$ layers. Here we present the first observations on the inter- and intra-layer phonon modes of $\mathrm{MnBi_{2n}Te_{3n+1}}$ (n=1,2,3,4) using cryogenic low-frequency Raman spectroscopy. We experimentally and theoretically distinguish the Raman vibrational modes using various polarization configurations. The two peaks at 66 cm$^{-1}$ and 112 cm$^{-1}$ show an abnormal perturbation in the Raman linewidths below the magnetic transition temperature due to spin-phonon coupling. In $\mathrm{MnBi_4Te_7}$, the $\mathrm{Bi_2Te_3}$ layers induce Davydov splitting of the A$_{1g}$ mode around 137 cm$^{-1}$ at 5 K. Using the linear chain model, we estimate the out-of-plane interlayer force constant to be $(3.98 \pm 0.14) \times 10^{19}$ N/m$^3$ at 5 K, three times weaker than that of $\mathrm{Bi_2Te_3}$. Our work discovers the dynamics of phonon modes of the $\mathrm{MnBi_2Te_4}$ and the effect of the additional $\mathrm{Bi_2Te_3}$ layers, providing the first-principles guidance to tailor the physical properties of layered heterostructures.
We propose a strategy for optimizing a sensor trajectory in order to estimate the time dependence of a localized scalar source in turbulent channel flow. The approach leverages the view of the adjoint scalar field as the sensitivity of measurement to a possible source. A cost functional is constructed so that the optimal sensor trajectory maintains a high sensitivity and low temporal variation in the measured signal, for a given source location. This naturally leads to the adjoint-of-adjoint equation based on which the sensor trajectory is iteratively optimized. It is shown that the estimation performance based on the measurement obtained by a sensor moving along the optimal trajectory is drastically improved from that achieved with a stationary sensor. It is also shown that the ratio of the fluctuation and the mean of the sensitivity for a given sensor trajectory can be used as a diagnostic tool to evaluate the resultant performance. Based on this finding, we propose a new cost functional which only includes the ratio without any adjustable parameters, and demonstrate its effectiveness in predicting the time dependence of scalar release from the source.
Gamma distributed delay differential equations (DDEs) arise naturally in many modelling applications. However, appropriate numerical methods for generic Gamma distributed DDEs are not currently available. Accordingly, modellers often resort to approximating the gamma distribution with an Erlang distribution and using the linear chain technique to derive an equivalent system of ordinary differential equations. In this work, we develop a functionally continuous Runge-Kutta method to numerically integrate the gamma distributed DDE and perform numerical tests to confirm the accuracy of the numerical method. As the functionally continuous Runge-Kutta method is not available in most scientific software packages, we then derive hypoexponential approximations of the gamma distributed DDE. Using our numerical method, we show that while using the common Erlang approximation can produce solutions that are qualitatively different from the underlying gamma distributed DDE, our hypoexponential approximations do not have this limitation. Finally, we implement our hypoexponential approximations to perform statistical inference on synthetic epidemiological data.
Markov state models (MSMs) have been broadly adopted for analyzing molecular dynamics trajectories, but the approximate nature of the models that results from coarse-graining into discrete states is a long-known limitation. We show theoretically that, despite the coarse graining, in principle MSM-like analysis can yield unbiased estimation of key observables. We describe unbiased estimators for equilibrium state populations, for the mean first-passage time (MFPT) of an arbitrary process, and for state committors - i.e., splitting probabilities. Generically, the estimators are only asymptotically unbiased but we describe how extension of a recently proposed reweighting scheme can accelerate relaxation to unbiased values. Exactly accounting for 'sliding window' averaging over finite-length trajectories is a key, novel element of our analysis. In general, our analysis indicates that coarse-grained MSMs are asymptotically unbiased for steady-state properties only when appropriate boundary conditions (e.g., source-sink for MFPT estimation) are applied directly to trajectories, prior to calculation of the appropriate transition matrix.
The phase field paradigm, in combination with a suitable variational structure, has opened a path for using Griffith's energy balance to predict the fracture of solids. These so-called phase field fracture methods have gained significant popularity over the past decade, and are now part of commercial finite element packages and engineering fitness-for-service assessments. Crack paths can be predicted, in arbitrary geometries and dimensions, based on a global energy minimisation - without the need for \textit{ad hoc} criteria. In this work, we review the fundamentals of phase field fracture methods and examine their capabilities in delivering predictions in agreement with the classical fracture mechanics theory pioneered by Griffith. The two most widely used phase field fracture models are implemented in the context of the finite element method, and several paradigmatic boundary value problems are addressed to gain insight into their predictive abilities across all cracking stages; both the initiation of growth and stable crack propagation are investigated. In addition, we examine the effectiveness of phase field models with an internal material length scale in capturing size effects and the transition flaw size concept. Our results show that phase field fracture methods satisfactorily approximate classical fracture mechanics predictions and can also reconcile stress and toughness criteria for fracture. The accuracy of the approximation is however dependent on modelling and constitutive choices; we provide a rationale for these differences and identify suitable approaches for delivering phase field fracture predictions that are in good agreement with well-established fracture mechanics paradigms.
The convergence property of a stochastic algorithm for the self-consistent field (SCF) calculations of electron structures is studied. The algorithm is formulated by rewriting the electron charges as a trace/diagonal of a matrix function, which is subsequently expressed as a statistical average. The function is further approximated by using a Krylov subspace approximation. As a result, each SCF iteration only samples one random vector without having to compute all the orbitals. We consider the common practice of SCF iterations with damping and mixing. We prove with appropriate assumptions that the iterations converge in the mean-square sense, when the stochastic error has an almost sure bound. We also consider the scenario when such an assumption is weakened to a second moment condition, and prove the convergence in probability.
Multiple organ failure (MOF) is a severe syndrome with a high mortality rate among Intensive Care Unit (ICU) patients. Early and precise detection is critical for clinicians to make timely decisions. An essential challenge in applying machine learning models to electronic health records (EHRs) is the pervasiveness of missing values. Most existing imputation methods are involved in the data preprocessing phase, failing to capture the relationship between data and outcome for downstream predictions. In this paper, we propose classifier-guided generative adversarial imputation networks Classifier-GAIN) for MOF prediction to bridge this gap, by incorporating both observed data and label information. Specifically, the classifier takes imputed values from the generator(imputer) to predict task outcomes and provides additional supervision signals to the generator by joint training. The classifier-guide generator imputes missing values with label-awareness during training, improving the classifier's performance during inference. We conduct extensive experiments showing that our approach consistently outperforms classical and state-of-art neural baselines across a range of missing data scenarios and evaluation metrics.
Instrumental systematics need to be controlled to high precision for upcoming Cosmic Microwave Background (CMB) experiments. The level of contamination caused by these systematics is often linked to the scan strategy, and scan strategies for satellite experiments can significantly mitigate these systematics. However, no detailed study has been performed for ground-based experiments. Here we show that under the assumption of constant elevation scans (CESs), the ability of the scan strategy to mitigate these systematics is strongly limited, irrespective of the detailed structure of the scan strategy. We calculate typical values and maps of the quantities coupling the scan to the systematics, and show how these quantities vary with the choice of observing elevations. These values and maps can be used to calculate and forecast the magnitude of different instrumental systematics without requiring detailed scan strategy simulations. As a reference point, we show that inclusion of even a single boresight rotation angle significantly improves over sky rotation alone for mitigating these systematics. A standard metric for evaluating cross-linking is related to one of the parameters studied in this work, so a corollary of our work is that the cross-linking will suffer from the same CES limitations and therefore upcoming CMB surveys will unavoidably have poorly cross-linked regions if they use CESs, regardless of detailed scheduling choices. Our results are also relevant for non-CMB surveys that perform constant elevation scans and may have scan-coupled systematics, such as intensity mapping surveys.
Integer quantization of neural networks can be defined as the approximation of the high precision computation of the canonical neural network formulation, using reduced integer precision. It plays a significant role in the efficient deployment and execution of machine learning (ML) systems, reducing memory consumption and leveraging typically faster computations. In this work, we present an integer-only quantization strategy for Long Short-Term Memory (LSTM) neural network topologies, which themselves are the foundation of many production ML systems. Our quantization strategy is accurate (e.g. works well with quantization post-training), efficient and fast to execute (utilizing 8 bit integer weights and mostly 8 bit activations), and is able to target a variety of hardware (by leveraging instructions sets available in common CPU architectures, as well as available neural accelerators).
Let ${\mathfrak M}=({\mathcal M},\rho)$ be a metric space and let $X$ be a Banach space. Let $F$ be a set-valued mapping from ${\mathcal M}$ into the family ${\mathcal K}_m(X)$ of all compact convex subsets of $X$ of dimension at most $m$. The main result in our recent joint paper with Charles Fefferman (which is referred to as a ``Finiteness Principle for Lipschitz selections'') provides efficient conditions for the existence of a Lipschitz selection of $F$, i.e., a Lipschitz mapping $f:{\mathcal M}\to X$ such that $f(x)\in F(x)$ for every $x\in{\mathcal M}$. We give new alternative proofs of this result in two special cases. When $m=2$ we prove it for $X={\bf R}^{2}$, and when $m=1$ we prove it for all choices of $X$. Both of these proofs make use of a simple reiteration formula for the ``core'' of a set-valued mapping $F$, i.e., for a mapping $G:{\mathcal M}\to{\mathcal K}_m(X)$ which is Lipschitz with respect to the Hausdorff distance, and such that $G(x)\subset F(x)$ for all $x\in{\mathcal M}$.
Purpose: Quantitative magnetization transfer (qMT) imaging can be used to quantify the proportion of protons in a voxel attached to macromolecules. Here, we show that the original qMT balanced steady-state free precession (bSSFP) model is biased due to over-simplistic assumptions made in its derivation. Theory and Methods: We present an improved model for qMT bSSFP, which incorporates finite radio-frequency (RF) pulse effects as well as simultaneous exchange and relaxation. Further, a correction to finite RF pulse effects for sinc-shaped excitations is derived. The new model is compared to the original one in numerical simulations of the Bloch-McConnell equations and in previously acquired in-vivo data. Results: Our numerical simulations show that the original signal equation is significantly biased in typical brain tissue structures (by 7-20 %) whereas the new signal equation outperforms the original one with minimal bias (< 1%). It is further shown that the bias of the original model strongly affects the acquired qMT parameters in human brain structures, with differences in the clinically relevant parameter of pool-size-ratio of up to 31 %. Particularly high biases of the original signal equation are expected in an MS lesion within diseased brain tissue (due to a low T2/T1-ratio), demanding a more accurate model for clinical applications. Conclusion: The improved model for qMT bSSFP is recommended for accurate qMT parameter mapping in healthy and diseased brain tissue structures.
We construct a categorical framework for nonlinear postquantum inference, with embeddings of convex closed sets of suitable reflexive Banach spaces as objects and pullbacks of Br\`egman quasi-nonexpansive mappings (in particular, constrained maximisations of Br\`egman relative entropies) as morphisms. It provides a nonlinear convex analytic analogue of Chencov's programme of geometric study of categories of linear positive maps between spaces of states, a working model of Mielnik's nonlinear transmitters, and a setting for nonlinear resource theories (with monoids of Br\`egman quasi-nonexpansive maps as free operations, their asymptotic fixed point sets as free sets, and Br\`egman relative entropies as resource monotones). We construct a range of concrete examples for semi-finite JBW-algebras and any W*-algebras. Due to relative entropy's asymmetry, all constructions have left and right versions, with Legendre duality inducing categorical equivalence between their well-defined restrictions. Inner groupoids of these categories implement the notion of statistical equivalence. The hom-sets of a subcategory of morphisms given by entropic projections have the structure of partially ordered commutative monoids (so, they are resource theories in Fritz's sense). Further restriction of objects to affine sets turns Br\`egman relative entropy into a functor. Finally, following Lawvere's adjointness paradigm for deductive logic, but with a semantic twist representing Jaynes' and Chencov's views on statistical inference, we introduce a category-theoretic multi-(co)agent setting for inductive inference theories, implemented by families of monads and comonads. We show that the br\`egmanian approach provides some special cases of this setting.
Garc\'ia-Aguilar et al. [Phys. Rev. Lett 126, 038001 (2021)] have shown that the deformations of "shape-shifting droplets" are consistent with an elastic model, that, unlike previous models, includes the intrinsic curvature of the frozen surfactant layer. In this Comment, we show that the interplay between surface tension and intrinsic curvature in their model is in fact mathematically equivalent to a physically very different phase-transition mechanism of the same process that we developed previously [Phys. Rev. Lett. 118, 088001 (2017); Phys. Rev. Res. 1, 023017 (2019)]. The mathematical models cannot therefore distinguish between the two mechanisms, and hence it is not possible to claim that one mechanism underlies all observed shape-shifting phenomena without a much more detailed comparison of experiment and theory.
We propose a novel scheme for the exact renormalisation group motivated by the desire of reducing the complexity of practical computations. The key idea is to specify renormalisation conditions for all inessential couplings, leaving us with the task of computing only the flow of the essential ones. To achieve this aim, we utilise a renormalisation group equation for the effective average action which incorporates general non-linear field reparameterisations. A prominent feature of the scheme is that, apart from the renormalisation of the mass, the propagator evaluated at any constant value of the field maintains its unrenormalised form. Conceptually, the scheme provides a clearer picture of renormalisation itself since the redundant, non-physical content is automatically disregarded in favour of a description based only on quantities that enter expressions for physical observables. To exemplify the scheme's utility, we investigate the Wilson-Fisher fixed point in three dimensions at order two in the derivative expansion. In this case, the scheme removes all order $\partial^2$ operators apart from the canonical term. Further simplifications occur at higher orders in the derivative expansion. Although we concentrate on a minimal scheme that reduces the complexity of computations, we propose more general schemes where inessential couplings can be tuned to optimise a given approximation. We further discuss the applicability of the scheme to a broad range of physical theories.
Amorphous dielectric materials have been known to host two-level systems (TLSs) for more than four decades. Recent developments on superconducting resonators and qubits enable detailed studies on the physics of TLSs. In particular, measuring the loss of a device over long time periods (a few days) allows us to investigate stochastic fluctuations due to the interaction between TLSs. We measure the energy relaxation time of a frequency-tunable planar superconducting qubit over time and frequency. The experiments show a variety of stochastic patterns that we are able to explain by means of extensive simulations. The model used in our simulations assumes a qubit interacting with high-frequency TLSs, which, in turn, interact with thermally activated low-frequency TLSs. Our simulations match the experiments and suggest the density of low-frequency TLSs is about three orders of magnitude larger than that of high-frequency ones.
Learning from implicit user feedback is challenging as we can only observe positive samples but never access negative ones. Most conventional methods cope with this issue by adopting a pairwise ranking approach with negative sampling. However, the pairwise ranking approach has a severe disadvantage in the convergence time owing to the quadratically increasing computational cost with respect to the sample size; it is problematic, particularly for large-scale datasets and complex models such as neural networks. By contrast, a pointwise approach does not directly solve a ranking problem, and is therefore inferior to a pairwise counterpart in top-K ranking tasks; however, it is generally advantageous in regards to the convergence time. This study aims to establish an approach to learn personalised ranking from implicit feedback, which reconciles the training efficiency of the pointwise approach and ranking effectiveness of the pairwise counterpart. The key idea is to estimate the ranking of items in a pointwise manner; we first reformulate the conventional pointwise approach based on density ratio estimation and then incorporate the essence of ranking-oriented approaches (e.g. the pairwise approach) into our formulation. Through experiments on three real-world datasets, we demonstrate that our approach not only dramatically reduces the convergence time (one to two orders of magnitude faster) but also significantly improving the ranking performance.
Full Duplex (FD) radio has emerged as a promising solution to increase the data rates by up to a factor of two via simultaneous transmission and reception in the same frequency band. This paper studies a novel hybrid beamforming (HYBF) design to maximize the weighted sum-rate (WSR) in a single-cell millimeter wave (mmWave) massive multiple-input-multiple-output (mMIMO) FD system. Motivated by practical considerations, we assume that the multi-antenna users and hybrid FD base station (BS) suffer from the limited dynamic range (LDR) noise due to non-ideal hardware and an impairment aware HYBF approach is adopted by integrating the traditional LDR noise model in the mmWave band. In contrast to the conventional HYBF schemes, our design also considers the joint sum-power and the practical per-antenna power constraints. A novel interference, self-interference (SI) and LDR noise aware optimal power allocation scheme for the uplink (UL) users and FD BS is also presented to satisfy the joint constraints. The maximum achievable gain of a multi-user mmWave FD system over a fully digital half duplex (HD) system with different LDR noise levels and numbers of the radio-frequency (RF) chains is investigated. Simulation results show that our design outperforms the HD system with only a few RF chains at any LDR noise level. The advantage of having amplitude control at the analog stage is also examined, and additional gain for the mmWave FD system becomes evident when the number of RF chains at the hybrid FD BS is small.
We study the influence of running vacuum on the baryon-to-photon ratio in running vacuum models (RVMs). When there exists a non-minimal coupling between photons and other matter in the expanding universe, the energy-momentum tensor of photons is no longer conserved, but the energy of photons could remain conserved. We discuss the conditions for the energy conservation of photons in RVMs. The photon number density and baryon number density, from the epoch of photon decoupling to the present day, are obtained in the context of RVMs by assuming that photons and baryons can be coupled to running vacuum, respectively. Both cases lead to a time-evolving baryon-to-photon ratio. However the evolution of the baryon-to-photon ratio is strictly constrained by observations. It is found that if the dynamic term of running vacuum is indeed coupled to photons or baryons, the coefficient of the dynamic term must be extremely small, which is unnatural. Therefore, our study basically rules out the possibility that running vacuum is coupled to photons or baryons in RVMs.
To extract liver from medical images is a challenging task due to similar intensity values of liver with adjacent organs, various contrast levels, various noise associated with medical images and irregular shape of liver. To address these issues, it is important to preprocess the medical images, i.e., computerized tomography (CT) and magnetic resonance imaging (MRI) data prior to liver analysis and quantification. This paper investigates the impact of permutation of various preprocessing techniques for CT images, on the automated liver segmentation using deep learning, i.e., U-Net architecture. The study focuses on Hounsfield Unit (HU) windowing, contrast limited adaptive histogram equalization (CLAHE), z-score normalization, median filtering and Block-Matching and 3D (BM3D) filtering. The segmented results show that combination of three techniques; HU-windowing, median filtering and z-score normalization achieve optimal performance with Dice coefficient of 96.93%, 90.77% and 90.84% for training, validation and testing respectively.
The Robot Operating System 2 (ROS2) targets distributed real-time systems and is widely used in the robotics community. Especially in these systems, latency in data processing and communication can lead to instabilities. Though being highly configurable with respect to latency, ROS2 is often used with its default settings. In this paper, we investigate the end-to-end latency of ROS2 for distributed systems with default settings and different Data Distribution Service (DDS) middlewares. In addition, we profile the ROS2 stack and point out latency bottlenecks. Our findings indicate that end-to-end latency strongly depends on the used DDS middleware. Moreover, we show that ROS2 can lead to 50% latency overhead compared to using low-level DDS communications. Our results imply guidelines for designing distributed ROS2 architectures and indicate possibilities for reducing the ROS2 overhead.
We study a long-recognised but under-appreciated symmetry called "dynamical similarity" and illustrate its relevance to many important conceptual problems in fundamental physics. Dynamical similarities are general transformations of a system where the unit of Hamilton's principal function is rescaled, and therefore represent a kind of dynamical scaling symmetry with formal properties that differ from many standard symmetries. To study this symmetry, we develop a general framework for symmetries that distinguishes the observable and surplus structures of a theory by using the minimal freely specifiable initial data for the theory that is necessary to achieve empirical adequacy. This framework is then applied to well-studied examples including Galilean invariance and the symmetries of the Kepler problem. We find that our framework gives a precise dynamical criterion for identifying the observables of those systems, and that those observables agree with epistemic expectations. We then apply our framework to dynamical similarity. First we give a general definition of dynamical similarity. Then we show, with the help of some previous results, how the dynamics of our observables leads to singularity resolution and the emergence of an arrow of time in cosmology.
State-space models (SSM) with Markov switching offer a powerful framework for detecting multiple regimes in time series, analyzing mutual dependence and dynamics within regimes, and asserting transitions between regimes. These models however present considerable computational challenges due to the exponential number of possible regime sequences to account for. In addition, high dimensionality of time series can hinder likelihood-based inference. This paper proposes novel statistical methods for Markov-switching SSMs using maximum likelihood estimation, Expectation-Maximization (EM), and parametric bootstrap. We develop solutions for initializing the EM algorithm, accelerating convergence, and conducting inference that are ideally suited to massive spatio-temporal data such as brain signals. We evaluate these methods in simulations and present applications to EEG studies of epilepsy and of motor imagery. All proposed methods are implemented in a MATLAB toolbox available at https://github.com/ddegras/switch-ssm.
An increasing amount of location-based service (LBS) data is being accumulated and helps to study urban dynamics and human mobility. GPS coordinates and other location indicators are normally low dimensional and only representing spatial proximity, thus difficult to be effectively utilized by machine learning models in Geo-aware applications. Existing location embedding methods are mostly tailored for specific problems that are taken place within areas of interest. When it comes to the scale of a city or even a country, existing approaches always suffer from extensive computational cost and significant data sparsity. Different from existing studies, we propose to learn representations through a GCN-aided skip-gram model named GCN-L2V by considering both spatial connection and human mobility. With a flow graph and a spatial graph, it embeds context information into vector representations. GCN-L2V is able to capture relationships among locations and provide a better notion of similarity in a spatial environment. Across quantitative experiments and case studies, we empirically demonstrate that representations learned by GCN-L2V are effective. As far as we know, this is the first study that provides a fine-grained location embedding at the city level using only LBS records. GCN-L2V is a general-purpose embedding model with high flexibility and can be applied in down-streaming Geo-aware applications.
The Earth's magnetotail is characterized by stretched magnetic field lines. Energetic particles are effectively scattered due to the field-line curvature, which then leads to isotropization of energetic particle distributions and particle precipitation to the Earth's atmosphere. Measurements of these precipitation at low-altitude spacecraft are thus often used to remotely probe the magnetotail current sheet configuration. This configuration may include spatially localized maxima of the curvature radius at equator (due to localized humps of the equatorial magnetic field magnitude) that reduce the energetic particle scattering and precipitation. Therefore, the measured precipitation patterns are related to the spatial distribution of the equatorial curvature radius that is determined by the magnetotail current sheet configuration. In this study, we show that, contrary to previous thoughts, the magnetic field line configuration with the localized curvature radius maximum can actually enhance the scattering and subsequent precipitation. The spatially localized magnetic field dipolarization (magnetic field humps) can significantly curve magnetic field lines far from the equator and create off-equatorial minima in the curvature radius. Scattering of energetic particles in these off-equatorial regions alters the scattering (and precipitation) patterns, which has not been studied yet. We discuss our results in the context of remote-sensing the magnetotail current sheet configuration with low-altitude spacecraft measurements.
Since the heavy neutrinos of the inverse seesaw mechanism mix largely with the standard ones, the charged currents formed with them and the muons have the potential of generating robust and positive contribution to the anomalous magnetic moment of the muon. Ho\-we\-ver, bounds from the non-unitary in the leptonic mixing matrix may restrict so severely the parameters of the mechanism that, depending on the framework under which the mechanism is implemented, may render it unable to explain the recent muon g-2 result. In this paper we show that this happens when we implement the mechanism into the standard model and into two versions of the 3-3-1 models.
Using an argument of Baldwin--Hu--Sivek, we prove that if $K$ is a hyperbolic fibered knot with fiber $F$ in a closed, oriented $3$--manifold $Y$, and $\widehat{HFK}(Y,K,[F], g(F)-1)$ has rank $1$, then the monodromy of $K$ is freely isotopic to a pseudo-Anosov map with no fixed points. In particular, this shows that the monodromy of a hyperbolic L-space knot is freely isotopic to a map with no fixed points.
The object of Weakly-supervised Temporal Action Localization (WS-TAL) is to localize all action instances in an untrimmed video with only video-level supervision. Due to the lack of frame-level annotations during training, current WS-TAL methods rely on attention mechanisms to localize the foreground snippets or frames that contribute to the video-level classification task. This strategy frequently confuse context with the actual action, in the localization result. Separating action and context is a core problem for precise WS-TAL, but it is very challenging and has been largely ignored in the literature. In this paper, we introduce an Action-Context Separation Network (ACSNet) that explicitly takes into account context for accurate action localization. It consists of two branches (i.e., the Foreground-Background branch and the Action-Context branch). The Foreground- Background branch first distinguishes foreground from background within the entire video while the Action-Context branch further separates the foreground as action and context. We associate video snippets with two latent components (i.e., a positive component and a negative component), and their different combinations can effectively characterize foreground, action and context. Furthermore, we introduce extended labels with auxiliary context categories to facilitate the learning of action-context separation. Experiments on THUMOS14 and ActivityNet v1.2/v1.3 datasets demonstrate the ACSNet outperforms existing state-of-the-art WS-TAL methods by a large margin.
Intrinsic Image Decomposition is an open problem of generating the constituents of an image. Generating reflectance and shading from a single image is a challenging task specifically when there is no ground truth. There is a lack of unsupervised learning approaches for decomposing an image into reflectance and shading using a single image. We propose a neural network architecture capable of this decomposition using physics-based parameters derived from the image. Through experimental results, we show that (a) the proposed methodology outperforms the existing deep learning-based IID techniques and (b) the derived parameters improve the efficacy significantly. We conclude with a closer analysis of the results (numerical and example images) showing several avenues for improvement.
We consider the barotropic Navier-Stokes system describing the motion of a compressible viscous fluid confined to a bounded domain driven by time periodic inflow/outflow boundary conditions. We show that the problem admits a time periodic solution in the class of weak solutions satisfying the energy inequality.
Understanding the features learned by deep models is important from a model trust perspective, especially as deep systems are deployed in the real world. Most recent approaches for deep feature understanding or model explanation focus on highlighting input data features that are relevant for classification decisions. In this work, we instead take the perspective of relating deep features to well-studied, hand-crafted features that are meaningful for the application of interest. We propose a methodology and set of systematic experiments for exploring deep features in this setting, where input feature importance approaches for deep feature understanding do not apply. Our experiments focus on understanding which hand-crafted and deep features are useful for the classification task of interest, how robust these features are for related tasks and how similar the deep features are to the meaningful hand-crafted features. Our proposed method is general to many application areas and we demonstrate its utility on orchestral music audio data.
An automorphism of a rooted spherically homogeneous tree is settled if it satisfies certain conditions on the growth of cycles at finite levels of the tree. In this paper, we consider a conjecture by Boston and Jones that the image of an arboreal representation of the absolute Galois group of a number field in the automorphism group of a tree has a dense subset of settled elements. Inspired by analogous notions in theory of compact Lie groups, we introduce the concepts of a maximal torus and a Weyl group for actions of profinite groups on rooted trees, and we show that the Weyl group contains important information about settled elements. We study maximal tori and their Weyl groups in the images of arboreal representations associated to quadratic polynomials over algebraic number fields, and in branch groups.
Open quantum systems exhibit a rich phenomenology, in comparison to closed quantum systems that evolve unitarily according to the Schr\"odinger equation. The dynamics of an open quantum system are typically classified into Markovian and non-Markovian, depending on whether the dynamics can be decomposed into valid quantum operations at any time scale. Since Markovian evolutions are easier to simulate, compared to non-Markovian dynamics, it is reasonable to assume that non-Markovianity can be employed for useful quantum-technological applications. Here, we demonstrate the usefulness of non-Markovianity for preserving correlations and coherence in quantum systems. For this, we consider a broad class of qubit evolutions, having a decoherence matrix separated from zero for large times. While any such Markovian evolution leads to an exponential loss of correlations, non-Markovianity can help to preserve correlations even in the limit $t \rightarrow \infty$. For covariant qubit evolutions, we also show that non-Markovianity can be used to preserve quantum coherence at all times, which is an important resource for quantum metrology. We explicitly demonstrate this effect experimentally with linear optics, by implementing the required evolution that is non-Markovian at all times.
People hope automated driving technology is always in a stable and controllable state; specifically, it can be divided into controllable planning, controllable responsibility, and controllable information. When this controllability is undermined, it brings about the problems, e.g., trolley dilemma, responsibility attribution, information leakage, and security. This article discusses these three types of issues separately and clarifies the misunderstandings.
In a recent paper with J.-P. Nicolas [J.-P. Nicolas and P.T. Xuan, Annales Henri Poincare 2019], we studied the peeling for scalar fields on Kerr metrics. The present work extends these results to Dirac fields on the same geometrical background. We follow the approach initiated by L.J. Mason and J.-P. Nicolas [L. Mason and J.-P. Nicolas, J.Inst.Math.Jussieu 2009; L. Mason and J.-P. Nicolas, J.Geom.Phys 2012] on the Schwarzschild spacetime and extended to Kerr metrics for scalar fields. The method combines the Penrose conformal compactification and geometric energy estimates in order to work out a definition of the peeling at all orders in terms of Sobolev regularity near $\mathscr{I}$, instead of ${\mathcal C}^k$ regularity at $\mathscr{I}$, then provides the optimal spaces of initial data such that the associated solution satisfies the peeling at a given order. The results confirm that the analogous decay and regularity assumptions on initial data in Minkowski and in Kerr produce the same regularity across null infinity. Our results are local near spacelike infinity and are valid for all values of the angular momentum of the spacetime, including for fast Kerr metrics.
Two-dimensional (2D) hybrid organic-inorganic perovskites (HOIPs) are introducing new directions in the 2D materials landscape. The coexistence of ferroelectricity and spin-orbit interactions play a key role in their optoelectronic properties. We perform a detailed study on a recently synthesized ferroelectric 2D-HOIP, (AMP)PbI$_4$ (AMP = 4-aminomethyl-piperidinium). The calculated polarization and Rashba parameter are in excellent agreement with experimental values. We report a striking new effect, i.e., an extraordinarily large Rashba anisotropy that is tunable by ferroelectric polarization: as polarization is reversed, not only the spin texture chirality is inverted, but also the major and minor axes of the Rashba anisotropy ellipse in k-space are interchanged - a pseudo rotation. A $k \cdot p$ model Hamiltonian and symmetry-mode analysis reveal a quadrilinear coupling between the cation-rotation modes responsible for the Rashba ellipse pseudo-rotation, the framework rotation, and the polarization. These findings may provide new avenues for spin-optoelectronic devices such as spin valves or spin FETs.
This paper presents a supervised learning method to generate continuous cost-to-go functions of non-holonomic systems directly from the workspace description. Supervision from informative examples reduces training time and improves network performance. The manifold representing the optimal trajectories of a non-holonomic system has high-curvature regions which can not be efficiently captured with uniform sampling. To address this challenge, we present an adaptive sampling method which makes use of sampling-based planners along with local, closed-form solutions to generate training samples. The cost-to-go function over a specific workspace is represented as a neural network whose weights are generated by a second, higher order network. The networks are trained in an end-to-end fashion. In our previous work, this architecture was shown to successfully learn to generate the cost-to-go functions of holonomic systems using uniform sampling. In this work, we show that uniform sampling fails for non-holonomic systems. However, with the proposed adaptive sampling methodology, our network can generate near-optimal trajectories for non-holonomic systems while avoiding obstacles. Experiments show that our method is two orders of magnitude faster compared to traditional approaches in cluttered environments.
We investigate the structure of the meson Regge trajectories based on the quadratic form of the spinless Salpeter-type equation. It is found that the forms of the Regge trajectories depend on the energy region. As the employed Regge trajectory formula does not match the energy region, the fitted parameters neither have explicit physical meanings nor obey the constraints although the fitted Regge trajectory can give the satisfactory predictions if the employed formula is appropriate mathematically. Moreover, the consistency of the Regge trajectories obtained from different approaches is discussed. And the Regge trajectories for different mesons are presented. Finally, we show that the masses of the constituents will come into the slope and explain why the slopes of the fitted linear Regge trajectories vary with different kinds of mesons.
Higgs-portal effective field theories are widely used as benchmarks in order to interpret collider and astroparticle searches for dark matter (DM) particles. To assess the validity of these effective models, it is important to confront them to concrete realizations that are complete in the ultraviolet regime. In this paper, we compare effective Higgs-portal models with scalar, fermionic and vector DM with a series of increasingly complex realistic models, taking into account all existing constraints from collider and astroparticle physics. These complete realizations include the inert doublet with scalar DM, the singlet-doublet model for fermionic DM and models based on spontaneously broken dark SU(2) and SU(3) gauge symmetries for vector boson DM. We also discuss the simpler scenarios in which a new scalar singlet field that mixes with the standard Higgs field is introduced with minimal couplings to isosinglet spin--$0, \frac12$ and 1 DM states. We show that in large regions of the parameter space of these models, the effective Higgs-portal approach provides a consistent limit and thus, can be safely adopted, in particular for the interpretation of searches for invisible Higgs boson decays at the LHC. The phenomenological implications of assuming or not that the DM states generate the correct cosmological relic density are also discussed.
We present geometric Bayesian active learning by disagreements (GBALD), a framework that performs BALD on its core-set construction interacting with model uncertainty estimation. Technically, GBALD constructs core-set on ellipsoid, not typical sphere, preventing low-representative elements from spherical boundaries. The improvements are twofold: 1) relieve uninformative prior and 2) reduce redundant estimations. Theoretically, geodesic search with ellipsoid can derive tighter lower bound on error and easier to achieve zero error than with sphere. Experiments show that GBALD has slight perturbations to noisy and repeated samples, and outperforms BALD, BatchBALD and other existing deep active learning approaches.
Facial Expression Recognition(FER) is one of the most important topic in Human-Computer interactions(HCI). In this work we report details and experimental results about a facial expression recognition method based on state-of-the-art methods. We fine-tuned a SeNet deep learning architecture pre-trained on the well-known VGGFace2 dataset, on the AffWild2 facial expression recognition dataset. The main goal of this work is to define a baseline for a novel method we are going to propose in the near future. This paper is also required by the Affective Behavior Analysis in-the-wild (ABAW) competition in order to evaluate on the test set this approach. The results reported here are on the validation set and are related on the Expression Challenge part (seven basic emotion recognition) of the competition. We will update them as soon as the actual results on the test set will be published on the leaderboard.
In this paper we compute the Newton polytope $\mathcal M_A$ of the Morse discriminant in the space of univariate polynomials with the given support set $A.$ Namely, we establish a surjection between the set of all combinatorial types of Morse univariate tropical polynomials and the vertices of $\mathcal M_A.$
Breast cancer is the most common invasive cancer in women, and the second main cause of death. Breast cancer screening is an efficient method to detect indeterminate breast lesions early. The common approaches of screening for women are tomosynthesis and mammography images. However, the traditional manual diagnosis requires an intense workload by pathologists, who are prone to diagnostic errors. Thus, the aim of this study is to build a deep convolutional neural network method for automatic detection, segmentation, and classification of breast lesions in mammography images. Based on deep learning the Mask-CNN (RoIAlign) method was developed to features selection and extraction; and the classification was carried out by DenseNet architecture. Finally, the precision and accuracy of the model is evaluated by cross validation matrix and AUC curve. To summarize, the findings of this study may provide a helpful to improve the diagnosis and efficiency in the automatic tumor localization through the medical image classification.
We have studied experimentally the generation of vortex flow by gravity waves with a frequency of 2.34 Hz excited on the water surface at an angle $2 \theta = arctan(3/4) \approx 36\deg$ to each other. The resulting horizontal surface flow has a stripe-like spatial structure. The width of the stripes L = $\pi$/(2ksin$\theta$) is determined by the wave vector k of the surface waves and the angle between them, and the length of the stripes is limited by the system size. It was found that the vertical vorticity $\Omega$ of the current on the fluid surface is proportional to the product of wave amplitudes, but its value is much higher than the value corresponding to the Stokes drift and it continues to grow with time even after the wave motion reaches a stationary regime. We demonstrate that the measured dependence $\Omega$(t) can be described within the recently developed model that takes into account the Eulerian contribution to the generated vortex flow and the effect of surface contamination. This model contains a free parameter that describes the elastic properties of the contaminated surface, and we also show that the found value of this parameter is in reasonable agreement with the measured decay rate of surface waves.
We investigate the impact of photochemical hazes and disequilibrium gases on the thermal structure of hot-Jupiters, using a detailed 1-D radiative-convective model. We find that the inclusion of photochemical hazes results in major heating of the upper and cooling of the lower atmosphere. Sulphur containing species, such as SH, S$_2$ and S$_3$ provide significant opacity in the middle atmosphere and lead to local heating near 1 mbar, while OH, CH, NH, and CN radicals produced by the photochemistry affect the thermal structure near 1 $\mu$bar. Furthermore we show that the modifications on the thermal structure from photochemical gases and hazes can have important ramifications for the interpretation of transit observations. Specifically, our study for the hazy HD 189733 b shows that the hotter upper atmosphere resulting from the inclusion of photochemical haze opacity imposes an expansion of the atmosphere, thus a steeper transit signature in the UV-Visible part of the spectrum. In addition, the temperature changes in the photosphere also affect the secondary eclipse spectrum. For HD 209458 b we find that a small haze opacity could be present in this atmosphere, at pressures below 1 mbar, which could be a result of both photochemical hazes and condensates. Our results motivate the inclusion of radiative feedback from photochemical hazes in general circulation models for a proper evaluation of atmospheric dynamics.
We compare the macroscopic and the local plastic behavior of a model amorphous solid based on two radically different numerical descriptions. On the one hand, we simulate glass samples by atomistic simulations. On the other, we implement a mesoscale elasto-plastic model based on a solid-mechanics description. The latter is extended to consider the anisotropy of the yield surface via statistically distributed local and discrete weak planes on which shear transformations can be activated. To make the comparison as quantitative as possible, we consider the simple case of a quasistatically driven two-dimensional system in the stationary flow state and compare mechanical observables measured on both models over the same length scales. We show that the macroscale response, including its fluctuations, can be quantitatively recovered for a range of elasto-plastic mesoscale parameters. Using a newly developed method that makes it possible to probe the local yield stresses in atomistic simulations, we calibrate the local mechanical response of the elasto-plastic model at different coarse-graining scales. In this case, the calibration shows a qualitative agreement only for an optimized subset of mesoscale parameters and for sufficiently coarse probing length scales. This calibration allows us to establish a length scale for the mesoscopic elements that corresponds to an upper bound of the shear transformation size, a key physical parameter in elasto-plastic models. We find that certain properties naturally emerge from the elasto-plastic model. In particular, we show that the elasto-plastic model reproduces the Bauschinger effect, namely the plasticity-induced anisotropy in the stress-strain response. We discuss the successes and failures of our approach, the impact of different model ingredients and propose future research directions for quantitative multi-scale models of amorphous plasticity.
This paper expounds very innovative results achieved between the mid-14th century and the beginning of the 16th century by Indian astronomers belonging to the so-called "M\=adhava school". These results were in keeping with researches in trigonometry: they concern the calculation of the eight of the circumference of a circle. They not only expose an analog of the series expansion of arctan(1) usually known as the "Leibniz series", but also other analogs of series expansions, the convergence of which is much faster. These series expansions are derived from evaluations of the rests of the partial sums of the primordial series, by means of some convergents of generalized continued fractions. A justification of these results in modern terms is provided, which aims at restoring their full mathematical interest.
Radio relics are the manifestation of electrons presumably being shock (re-)accelerated to high energies in the outskirts of galaxy clusters. However, estimates of the shocks' strength yield different results when measured with radio or X-ray observations. In general, Mach numbers obtained from radio observations are larger than the corresponding X-ray measurements. In this work, we investigate this Mach number discrepancy. For this purpose, we used the cosmological code ENZO to simulate a sample of galaxy clusters that host bright radio relics. For each relic, we computed the radio Mach number from the integrated radio spectrum and the X-ray Mach number from the X-ray surface brightness and temperature jumps. Our analysis suggests that the differences in the Mach number estimates follow from the way in which different observables are related to different parts of the underlying Mach number distribution: radio observations are more sensistive to the high Mach numbers present only in a small fraction of a shock's surface, while X-ray measurements reflect the average of the Mach number distribution. Moreover, X-ray measurements are very sensitive to the relic's orientation. If the same relic is observed from different sides, the measured X-ray Mach number varies significantly. On the other hand, the radio measurements are more robust, as they are unaffected by the relic's orientation.
In this paper, we present a sharp analysis for an alternating gradient descent algorithm which is used to solve the covariate adjusted precision matrix estimation problem in the high dimensional setting. Without the resampling assumption, we demonstrate that this algorithm not only enjoys a linear rate of convergence, but also attains the optimal statistical rate (i.e., minimax rate). Moreover, our analysis also characterizes the time-data tradeoffs in the covariate adjusted precision matrix estimation problem. Numerical experiments are provided to verify our theoretical results.
FISTA is a popular convex optimisation algorithm which is known to converge at an optimal rate whenever the optimisation domain is contained in a suitable Hilbert space. We propose a modified algorithm where each iteration is performed in a subspace, and that subspace is allowed to change at every iteration. Analytically, this allows us to guarantee convergence in a Banach space setting, although at a reduced rate depending on the conditioning of the specific problem. Numerically we show that a greedy adaptive choice of discretisation can greatly increase the time and memory efficiency in infinite dimensional Lasso optimisation problems.
When two identical two-dimensional (2D) periodic lattices are stacked in parallel after rotating one layer by a certain angle relative to the other layer, the resulting bilayer system can lose lattice periodicity completely and become a 2D quasicrystal. Twisted bilayer graphene with 30-degree rotation is a representative example. We show that such quasicrystalline bilayer systems generally develop macroscopically degenerate localized zero-energy states (ZESs) in strong coupling limit where the interlayer couplings are overwhelmingly larger than the intralayer couplings. The emergent chiral symmetry in strong coupling limit and aperiodicity of bilayer quasicrystals guarantee the existence of the ZESs. The macroscopically degenerate ZESs are analogous to the flat bands of periodic systems, in that both are composed of localized eigenstates, which give divergent density of states. For monolayers, we consider the triangular, square, and honeycomb lattices, comprised of homogenous tiling of three possible planar regular polygons: the equilateral triangle, square, and regular hexagon. We construct a compact theoretical framework, which we call the quasiband model, that describes the low energy properties of bilayer quasicrystals and counts the number of ZESs using a subset of Bloch states of monolayers. We also propose a simple geometric scheme in real space which can show the spatial localization of ZESs and count their number. Our work clearly demonstrates that bilayer quasicrystals in strong coupling limit are an ideal playground to study the intriguing interplay of flat band physics and the aperiodicity of quasicrystals.
In this paper, we delve into semi-supervised object detection where unlabeled images are leveraged to break through the upper bound of fully-supervised object detection models. Previous semi-supervised methods based on pseudo labels are severely degenerated by noise and prone to overfit to noisy labels, thus are deficient in learning different unlabeled knowledge well. To address this issue, we propose a data-uncertainty guided multi-phase learning method for semi-supervised object detection. We comprehensively consider divergent types of unlabeled images according to their difficulty levels, utilize them in different phases and ensemble models from different phases together to generate ultimate results. Image uncertainty guided easy data selection and region uncertainty guided RoI Re-weighting are involved in multi-phase learning and enable the detector to concentrate on more certain knowledge. Through extensive experiments on PASCAL VOC and MS COCO, we demonstrate that our method behaves extraordinarily compared to baseline approaches and outperforms them by a large margin, more than 3% on VOC and 2% on COCO.
This paper theoretically investigates the following empirical phenomenon: given a high-complexity network with poor generalization bounds, one can distill it into a network with nearly identical predictions but low complexity and vastly smaller generalization bounds. The main contribution is an analysis showing that the original network inherits this good generalization bound from its distillation, assuming the use of well-behaved data augmentation. This bound is presented both in an abstract and in a concrete form, the latter complemented by a reduction technique to handle modern computation graphs featuring convolutional layers, fully-connected layers, and skip connections, to name a few. To round out the story, a (looser) classical uniform convergence analysis of compression is also presented, as well as a variety of experiments on cifar and mnist demonstrating similar generalization performance between the original network and its distillation.
Distributed networks and real-time systems are becoming the most important components for the new computer age, the Internet of Things (IoT), with huge data streams or data sets generated from sensors and data generated from existing legacy systems. The data generated offers the ability to measure, infer and understand environmental indicators, from delicate ecologies and natural resources to urban environments. This can be achieved through the analysis of the heterogeneous data sources (structured and unstructured). In this paper, we propose a distributed framework Event STream Processing Engine for Environmental Monitoring Domain (ESTemd) for the application of stream processing on heterogeneous environmental data. Our work in this area demonstrates the useful role big data techniques can play in an environmental decision support system, early warning and forecasting systems. The proposed framework addresses the challenges of data heterogeneity from heterogeneous systems and real time processing of huge environmental datasets through a publish/subscribe method via a unified data pipeline with the application of Apache Kafka for real time analytics.
Morphological Segmentation involves decomposing words into morphemes, the smallest meaning-bearing units of language. This is an important NLP task for morphologically-rich agglutinative languages such as the Southern African Nguni language group. In this paper, we investigate supervised and unsupervised models for two variants of morphological segmentation: canonical and surface segmentation. We train sequence-to-sequence models for canonical segmentation, where the underlying morphemes may not be equal to the surface form of the word, and Conditional Random Fields (CRF) for surface segmentation. Transformers outperform LSTMs with attention on canonical segmentation, obtaining an average F1 score of 72.5% across 4 languages. Feature-based CRFs outperform bidirectional LSTM-CRFs to obtain an average of 97.1% F1 on surface segmentation. In the unsupervised setting, an entropy-based approach using a character-level LSTM language model fails to outperforms a Morfessor baseline, while on some of the languages neither approach performs much better than a random baseline. We hope that the high performance of the supervised segmentation models will help to facilitate the development of better NLP tools for Nguni languages.
We propose and study a new mathematical model of the human immunodeficiency virus (HIV). The main novelty is to consider that the antibody growth depends not only on the virus and on the antibodies concentration but also on the uninfected cells concentration. The model consists of five nonlinear differential equations describing the evolution of the uninfected cells, the infected ones, the free viruses, and the adaptive immunity. The adaptive immune response is represented by the cytotoxic T-lymphocytes (CTL) cells and the antibodies with the growth function supposed to be trilinear. The model includes two kinds of treatments. The objective of the first one is to reduce the number of infected cells, while the aim of the second is to block free viruses. Firstly, the positivity and the boundedness of solutions are established. After that, the local stability of the disease free steady state and the infection steady states are characterized. Next, an optimal control problem is posed and investigated. Finally, numerical simulations are performed in order to show the behavior of solutions and the effectiveness of the two incorporated treatments via an efficient optimal control strategy.
We use high quality VLT/MUSE data to study the kinematics and the ionized gas properties of Haro 11, a well known starburst merger system and the closest confirmed Lyman continuum leaking galaxy. We present results from integrated line maps, and from maps in three velocity bins comprising the blueshifted, systemic and redshifted emission. The kinematic analysis reveals complex velocities resulting from the interplay of virial motions and momentum feedback. Star formation happens intensively in three compact knots (knots A, B and C), but one, knot C, dominates the energy released in supernovae. The halo is characterised by low gas density and extinction, but with large temperature variations, coincident with fast shock regions. Moreover, we find large temperature discrepancies in knot C, when using different temperature-sensitive lines. The relative impact of the knots in the metal enrichment differs. While knot B is strongly enriching its closest surrounding, knot C is likely the main distributor of metals in the halo. In knot A, part of the metal enriched gas seems to escape through low density channels towards the south. We compare the metallicities from two methods and find large discrepancies in knot C, a shocked area, and the highly ionized zones, that we partially attribute to the effect of shocks. This work shows, that traditional relations developed from averaged measurements or simplified methods, fail to probe the diverse conditions of the gas in extreme environments. We need robust relations that include realistic models where several physical processes are simultaneously at work.
We demonstrate a method that merges the quantum filter diagonalization (QFD) approach for hybrid quantum/classical solution of the time-independent electronic Schr\"odinger equation with a low-rank double factorization (DF) approach for the representation of the electronic Hamiltonian. In particular, we explore the use of sparse "compressed" double factorization (C-DF) truncation of the Hamiltonian within the time-propagation elements of QFD, while retaining a similarly compressed but numerically converged double-factorized representation of the Hamiltonian for the operator expectation values needed in the QFD quantum matrix elements. Together with significant circuit reduction optimizations and number-preserving post-selection/echo-sequencing error mitigation strategies, the method is found to provide accurate predictions for low-lying eigenspectra in a number of representative molecular systems, while requiring reasonably short circuit depths and modest measurement costs. The method is demonstrated by experiments on noise-free simulators, decoherence- and shot-noise including simulators, and real quantum hardware.
The Android operating system is the most spread mobile platform in the world. Therefor attackers are producing an incredible number of malware applications for Android. Our aim is to detect Android's malware in order to protect the user. To do so really good results are obtained by dynamic analysis of software, but it requires complex environments. In order to achieve the same level of precision we analyze the machine code and investigate the frequencies of ngrams of opcodes in order to detect singular code blocks. This allow us to construct a database of infected code blocks. Then, because attacker may modify and organized differently the infected injected code in their new malware, we perform not only a semantic comparison of the tested software with the database of infected code blocks but also a structured comparison. To do such comparison we compute subgraph isomorphism. It allows us to characterize precisely if the tested software is a malware and if so in witch family it belongs. Our method is tested both on a laboratory database and a set of real data. It achieves an almost perfect detection rate.
This thesis is concerned with continuous, static, and single-objective optimization problems subject to inequality constraints. Nevertheless, some methods to handle other kinds of problems are briefly reviewed. The particle swarm optimization paradigm was inspired by previous simulations of the cooperative behaviour observed in social beings. It is a bottom-up, randomly weighted, population-based method whose ability to optimize emerges from local, individual-to-individual interactions. As opposed to traditional methods, it can deal with different problems with few or no adaptation due to the fact that it does profit from problem-specific features of the problem at issue but performs a parallel, cooperative exploration of the search-space by means of a population of individuals. The main goal of this thesis consists of developing an optimizer that can perform reasonably well on most problems. Hence, the influence of the settings of the algorithm's parameters on the behaviour of the system is studied, some general-purpose settings are sought, and some variations to the canonical version are proposed aiming to turn it into a more general-purpose optimizer. Since no termination condition is included in the canonical version, this thesis is also concerned with the design of some stopping criteria which allow the iterative search to be terminated if further significant improvement is unlikely, or if a certain number of time-steps are reached. In addition, some constraint-handling techniques are incorporated into the canonical algorithm to handle inequality constraints. Finally, the capabilities of the proposed general-purpose optimizers are illustrated by optimizing a few benchmark problems.
In recent years, human activity recognition has garnered considerable attention both in industrial and academic research because of the wide deployment of sensors, such as accelerometers and gyroscopes, in products such as smartphones and smartwatches. Activity recognition is currently applied in various fields where valuable information about an individual's functional ability and lifestyle is needed. In this study, we used the popular WISDM dataset for activity recognition. Using multivariate analysis of covariance (MANCOVA), we established a statistically significant difference (p<0.05) between the data generated from the sensors embedded in smartphones and smartwatches. By doing this, we show that smartphones and smartwatches don't capture data in the same way due to the location where they are worn. We deployed several neural network architectures to classify 15 different hand and non-hand-oriented activities. These models include Long short-term memory (LSTM), Bi-directional Long short-term memory (BiLSTM), Convolutional Neural Network (CNN), and Convolutional LSTM (ConvLSTM). The developed models performed best with watch accelerometer data. Also, we saw that the classification precision obtained with the convolutional input classifiers (CNN and ConvLSTM) was higher than the end-to-end LSTM classifier in 12 of the 15 activities. Additionally, the CNN model for the watch accelerometer was better able to classify non-hand oriented activities when compared to hand-oriented activities.
A space-time Trefftz discontinuous Galerkin method for the Schr\"odinger equation with piecewise-constant potential is proposed and analyzed. Following the spirit of Trefftz methods, trial and test spaces are spanned by non-polynomial complex wave functions that satisfy the Schro\"odinger equation locally on each element of the space-time mesh. This allows for a significant reduction in the number of degrees of freedom in comparison with full polynomial spaces. We prove well-posedness and stability of the method, and, for the one- and two- dimensional cases, optimal, high-order, h-convergence error estimates in a skeleton norm. Some numerical experiments validate the theoretical results presented.
We demonstrate the use of multiple atomic-level Rydberg-atom schemes for continuous frequency detection of radio frequency (RF) fields. Resonant detection of RF fields by electromagnetically-induced transparency and Autler-Townes (AT) in Rydberg atoms is typically limited to frequencies within the narrow bandwidth of a Rydberg transition. By applying a second field resonant with an adjacent Rydberg transition, far-detuned fields can be detected through a two-photon resonance AT splitting. This two-photon AT splitting method is several orders of magnitude more sensitive than off-resonant detection using the Stark shift. We present the results of various experimental configurations and a theoretical analysis to illustrate the effectiveness of this multiple level scheme. These results show that this approach allows for the detection of frequencies in continuous band between resonances with adjacent Rydberg states.
We study the space of $C^1$ isogeometric spline functions defined on trilinearly parameterized multi-patch volumes. Amongst others, we present a general framework for the design of the $C^1$ isogeometric spline space and of an associated basis, which is based on the two-patch construction [7], and which works uniformly for any possible multi-patch configuration. The presented method is demonstrated in more detail on the basis of a particular subclass of trilinear multi-patch volumes, namely for the class of trilinearly parameterized multi-patch volumes with exactly one inner edge. For this specific subclass of trivariate multi-patch parameterizations, we further numerically compute the dimension of the resulting $C^1$ isogeometric spline space and use the constructed $C^1$ isogeometric basis functions to numerically explore the approximation properties of the $C^1$ spline space by performing $L^2$ approximation.
Recently, higher-order topological matter and 3D quantum Hall effects have attracted great attention. The Fermi-arc mechanism of the 3D quantum Hall effect proposed in Weyl semimetals is characterized by the one-sided hinge states, which do not exist in all the previous quantum Hall systems and more importantly pose a realistic example of the higher-order topological matter. The experimental effort so far is in the Dirac semimetal Cd$_3$As$_2$, where however time-reversal symmetry leads to hinge states on both sides of the top/bottom surfaces, instead of the aspired one-sided hinge states. We propose that under a tilted magnetic field, the hinge states in Cd$_3$As$_2$-like Dirac semimetals can be one-sided, highly tunable by field direction and Fermi energy, and robust against weak disorder. Furthermore, we propose a scanning tunneling Hall measurement to detect the one-sided hinge states. Our results will be insightful for exploring not only the quantum Hall effects beyond two dimensions, but also other higher-order topological insulators in the future.
When two spherical particles submerged in a viscous fluid are subjected to an oscillatory flow, they align themselves perpendicular to the direction of the flow leaving a small gap between them. The formation of this compact structure is attributed to a non-zero residual flow known as steady streaming. We have performed direct numerical simulations of a fully-resolved, oscillating flow in which the pair of particles is modeled using an immersed boundary method. Our simulations show that the particles oscillate both parallel and perpendicular to the oscillating flow in elongated figure 8 trajectories. In absence of bottom friction, the mean gap between the particles depends only on the normalized Stokes boundary layer thickness $\delta^*$, and on the normalized, streamwise excursion length of the particles relative to the fluid $A_r^*$ (equivalent to the Keulegan-Carpenter number). For $A_r^*\lesssim 1$, viscous effects dominate and the mean particle separation only depends on $\delta^*$. For larger $A_r^*$-values, advection becomes important and the gap widens. Overall, the normalized mean gap between the particles scales as $L^*\approx3.0{\delta^*}^{1.5}+0.03{A_r^*}^3$, which also agrees well with previous experimental results. The two regimes are also observed in the magnitude of the oscillations of the gap perpendicular to the flow, which increases in the viscous regime and decreases in the advective regime. When bottom friction is considered, particle rotation increases and the gap widens. Our results stress the importance of simulating the particle motion with all its degrees of freedom to accurately model the system and reproduce experimental results. The new insights of the particle pairs provide an important step towards understanding denser and more complex systems.
We present a new model to describe the star formation process in galaxies, which includes the description of the different gas phases -- molecular, atomic, and ionized -- together with its metal content. The model, which will be coupled to cosmological simulations of galaxy formation, will be used to investigate the relation between the star formation rate (SFR) and the formation of molecular hydrogen. The model follows the time evolution of the molecular, atomic and ionized phases in a gas cloud and estimates the amount of stellar mass formed, by solving a set of five coupled differential equations. As expected, we find a positive, strong correlation between the molecular fraction and the initial gas density, which manifests in a positive correlation between the initial gas density and the SFR of the cloud.
The development of lightweight object detectors is essential due to the limited computation resources. To reduce the computation cost, how to generate redundant features plays a significant role. This paper proposes a new lightweight Convolution method Cross-Stage Lightweight (CSL) Module, to generate redundant features from cheap operations. In the intermediate expansion stage, we replaced Pointwise Convolution with Depthwise Convolution to produce candidate features. The proposed CSL-Module can reduce the computation cost significantly. Experiments conducted at MS-COCO show that the proposed CSL-Module can approximate the fitting ability of Convolution-3x3. Finally, we use the module to construct a lightweight detector CSL-YOLO, achieving better detection performance with only 43% FLOPs and 52% parameters than Tiny-YOLOv4.
The dispersion of a tracer in a fluid flow is influenced by the Lagrangian motion of fluid elements. Even in laminar regimes, the irregular chaotic behavior of a fluid flow can lead to effective stirring that rapidly redistributes a tracer throughout the domain. When the advected particles possess a finite size and nontrivial shape, however, their dynamics can differ markedly from passive tracers, thus affecting the dispersion phenomena. Here we investigate the behavior of neutrally buoyant particles in 2-dimensional chaotic flows, combining numerical simulations and laboratory experiments. We show that depending on the particles shape and size, the underlying Lagrangian coherent structures can be altered, resulting in distinct dispersion phenomena within the same flow field. Experiments performed in a two-dimensional cellular flow, exhibited a focusing effect in vortex cores of particles with anisotropic shape. In agreement with our numerical model, neutrally buoyant ellipsoidal particles display markedly different trajectories and overall organization than spherical particles, with a clustering in vortices that changes accordingly with the aspect ratio of the particles.
We explore the ability of overparameterized shallow neural networks to learn Lipschitz regression functions with and without label noise when trained by Gradient Descent (GD). To avoid the problem that in the presence of noisy labels, neural networks trained to nearly zero training error are inconsistent on this class, we propose an early stopping rule that allows us to show optimal rates. This provides an alternative to the result of Hu et al. (2021) who studied the performance of $\ell 2$ -regularized GD for training shallow networks in nonparametric regression which fully relied on the infinite-width network (Neural Tangent Kernel (NTK)) approximation. Here we present a simpler analysis which is based on a partitioning argument of the input space (as in the case of 1-nearest-neighbor rule) coupled with the fact that trained neural networks are smooth with respect to their inputs when trained by GD. In the noise-free case the proof does not rely on any kernelization and can be regarded as a finite-width result. In the case of label noise, by slightly modifying the proof, the noise is controlled using a technique of Yao, Rosasco, and Caponnetto (2007).
Pretrained language models have significantly improved the performance of down-stream language understanding tasks, including extractive question answering, by providing high-quality contextualized word embeddings. However, learning question answering models still need large-scaled data annotation in specific domains. In this work, we propose a cooperative, self-play learning framework, REGEX, for question generation and answering. REGEX is built upon a masked answer extraction task with an interactive learning environment containing an answer entity REcognizer, a question Generator, and an answer EXtractor. Given a passage with a masked entity, the generator generates a question around the entity, and the extractor is trained to extract the masked entity with the generated question and raw texts. The framework allows the training of question generation and answering models on any text corpora without annotation. We further leverage a reinforcement learning technique to reward generating high-quality questions and to improve the answer extraction model's performance. Experiment results show that REGEX outperforms the state-of-the-art (SOTA) pretrained language models and zero-shot approaches on standard question-answering benchmarks, and yields the new SOTA performance under the zero-shot setting.
Enterprise knowledge is a key asset in the competing and fast-changing corporate landscape. The ability to learn, store and distribute implicit and explicit knowledge can be the difference between success and failure. While enterprise knowledge management is a well-defined research domain, current implementations lack orientation towards small and medium enterprise. We propose a semantic search engine for relevant documents in an enterprise, based on automatic generated domain ontologies. In this paper we focus on the component for ontology learning and population.
Recently, Doroudiani and Karimipour [Phys. Rev. A \textbf{102} 012427(2020)] proposed the notation of planar maximally entangled (PME) states which are a wider class of multipartite entangled states than absolutely maximally entangled (AME) states. There they presented their constructions in the multipartite systems but the number of particles is restricted to be even. Here we first solve the remaining cases, i.e., constructions of planar maximally entangled states on systems with odd number of particles. In addition, we generalized the PME to the planar $k$-uniform states whose reductions to any adjacent $k$ parties along a circle of $N$ parties are maximally mixed. We presented a method to construct sets of planar $k$-uniform states which have minimal support.
We construct exact solutions to the Einstein-Maxwell theory with uplifting the four dimensional Fubini-Study Kahler manifold. We find the solutions can be expressed exactly as the integrals of two special functions. The solutions are regular almost everywhere except a bolt structure on a single point in any dimensionality. We also show that the solutions are unique and can not be non-trivially extended to include the cosmological constant in any dimensions.
The introduction of an optical resonator can enable efficient and precise interaction between a photon and a solid-state emitter. It facilitates the study of strong light-matter interaction, polaritonic physics and presents a powerful interface for quantum communication and computing. A pivotal aspect in the progress of light-matter interaction with solid-state systems is the challenge of combining the requirements of cryogenic temperature and high mechanical stability against vibrations while maintaining sufficient degrees of freedom for in-situ tunability. Here, we present a fiber-based open Fabry-P\'{e}rot cavity in a closed-cycle cryostat exhibiting ultra-high mechanical stability while providing wide-range tunability in all three spatial directions. We characterize the setup and demonstrate the operation with the root-mean-square cavity length fluctuation of less than $90$ pm at temperature of $6.5$ K and integration bandwidth of $100$ kHz. Finally, we benchmark the cavity performance by demonstrating the strong-coupling formation of exciton-polaritons in monolayer WSe$_2$ with a cooperativity of $1.6$. This set of results manifests the open-cavity in a closed-cycle cryostat as a versatile and powerful platform for low-temperature cavity QED experiments.
We determine the dark matter pair-wise relative velocity distribution in a set of Milky Way-like halos in the Auriga and APOSTLE simulations. Focusing on the smooth halo component, the relative velocity distribution is well-described by a Maxwell-Boltzmann distribution over nearly all radii in the halo. We explore the implications for velocity-dependent dark matter annihilation, focusing on four models which scale as different powers of the relative velocity: Sommerfeld, s-wave, p-wave, and d-wave models. We show that the J-factors scale as the moments of the relative velocity distribution, and that the halo-to-halo scatter is largest for d-wave, and smallest for Sommerfeld models. The J-factor is strongly correlated with the dark matter density in the halo, and is very weakly correlated with the velocity dispersion. This implies that if the dark matter density in the Milky Way can be robustly determined, one can accurately predict the dark matter annihilation signal, without the need to identify the dark matter velocity distribution in the Galaxy.
In this essay, we qualitatively demonstrate how small non-perturbative corrections are a necessary addition to semiclassical gravity's path integral. We use this to discuss implications for Hawking's information paradox and the bags of gold paradox.
We report Keck-NIRSPEC observations of the Brackett $\alpha$ 4.05 $\mu$m recombination line across the two candidate embedded super star clusters (SSCs) in NGC 1569. These SSCs power a bright HII region and have been previously detected as radio and mid-infrared sources. Supplemented with high resolution VLA mapping of the radio continuum along with IRTF-TEXES spectroscopy of the [SIV] 10.5 $\mu$m line, the Brackett $\alpha$ spectra data provide new insight into the dynamical state of gas ionized by these forming massive clusters. NIR sources detected in 2 $\mu$m images from the Slit-viewing Camera are matched with GAIA sources to obtain accurate celestial coordinates and slit positions to within $\sim 0.1''$. Br$\alpha$ is detected as a strong emission peak powered by the less luminous infrared source, MIR1 ($L_{\rm IR}\sim 2\times10^7~L_\odot$). The second candidate SSC MIR2 is more luminous ($L_{\rm IR}\gtrsim 4\times10^8~L_\odot$) but exhibits weak radio continuum and Br$\alpha$ emission, suggesting the ionized gas is extremely dense ($n_e\gtrsim 10^5$ cm$^{-3}$), corresponding to hypercompact HII regions around newborn massive stars. The Br$\alpha$ and [SIV] lines across the region are both remarkably symmetric and extremely narrow, with observed line widths $\Delta v \simeq 40$ km s$^{-1}$, FWHM. This result is the first clear evidence that feedback from NGC 1569's youngest giant clusters is currently incapable of rapid gas dispersal, consistent with the emerging theoretical paradigm in the formation of giant star clusters.
Machine learning has brought striking advances in multilingual natural language processing capabilities over the past year. For example, the latest techniques have improved the state-of-the-art performance on the XTREME multilingual benchmark by more than 13 points. While a sizeable gap to human-level performance remains, improvements have been easier to achieve in some tasks than in others. This paper analyzes the current state of cross-lingual transfer learning and summarizes some lessons learned. In order to catalyze meaningful progress, we extend XTREME to XTREME-R, which consists of an improved set of ten natural language understanding tasks, including challenging language-agnostic retrieval tasks, and covers 50 typologically diverse languages. In addition, we provide a massively multilingual diagnostic suite (MultiCheckList) and fine-grained multi-dataset evaluation capabilities through an interactive public leaderboard to gain a better understanding of such models. The leaderboard and code for XTREME-R will be made available at https://sites.research.google/xtreme and https://github.com/google-research/xtreme respectively.
This paper presents a state-of-the-art LiDAR based autonomous navigation system for under-canopy agricultural robots. Under-canopy agricultural navigation has been a challenging problem because GNSS and other positioning sensors are prone to significant errors due to attentuation and multi-path caused by crop leaves and stems. Reactive navigation by detecting crop rows using LiDAR measurements is a better alternative to GPS but suffers from challenges due to occlusion from leaves under the canopy. Our system addresses this challenge by fusing IMU and LiDAR measurements using an Extended Kalman Filter framework on low-cost hardwware. In addition, a local goal generator is introduced to provide locally optimal reference trajectories to the onboard controller. Our system is validated extensively in real-world field environments over a distance of 50.88~km on multiple robots in different field conditions across different locations. We report state-of-the-art distance between intervention results, showing that our system is able to safely navigate without interventions for 386.9~m on average in fields without significant gaps in the crop rows, 56.1~m in production fields and 47.5~m in fields with gaps (space of 1~m without plants in both sides of the row).
Drone imagery is increasingly used in automated inspection for infrastructure surface defects, especially in hazardous or unreachable environments. In machine vision, the key to crack detection rests with robust and accurate algorithms for image processing. To this end, this paper proposes a deep learning approach using hierarchical convolutional neural networks with feature preservation (HCNNFP) and an intercontrast iterative thresholding algorithm for image binarization. First, a set of branch networks is proposed, wherein the output of previous convolutional blocks is half-sizedly concatenated to the current ones to reduce the obscuration in the down-sampling stage taking into account the overall information loss. Next, to extract the feature map generated from the enhanced HCNN, a binary contrast-based autotuned thresholding (CBAT) approach is developed at the post-processing step, where patterns of interest are clustered within the probability map of the identified features. The proposed technique is then applied to identify surface cracks on the surface of roads, bridges or pavements. An extensive comparison with existing techniques is conducted on various datasets and subject to a number of evaluation criteria including the average F-measure (AF\b{eta}) introduced here for dynamic quantification of the performance. Experiments on crack images, including those captured by unmanned aerial vehicles inspecting a monorail bridge. The proposed technique outperforms the existing methods on various tested datasets especially for GAPs dataset with an increase of about 1.4% in terms of AF\b{eta} while the mean percentage error drops by 2.2%. Such performance demonstrates the merits of the proposed HCNNFP architecture for surface defect inspection.
We present a framework for simulating realistic inverse synthetic aperture radar images of automotive targets at millimeter wave frequencies. The model incorporates radar scattering phenomenology of commonly found vehicles along with range-Doppler based clutter and receiver noise. These images provide insights into the physical dimensions of the target, the number of wheels and the trajectory undertaken by the target. The model is experimentally validated with measurement data gathered from an automotive radar. The images from the simulation database are subsequently classified using both traditional machine learning techniques as well as deep neural networks based on transfer learning. We show that the ISAR images offer a classification accuracy above 90% and are robust to both noise and clutter.
There are two cases when the nonlinear Schr\"odinger equation (NLSE) with an external complex potential is well-known to support continuous families of localized stationary modes: the ${\cal PT}$-symmetric potentials and the Wadati potentials. Recently Y. Kominis and coauthors [Chaos, Solitons and Fractals, 118, 222-233 (2019)] have suggested that the continuous families can be also found in complex potentials of the form $W(x)=W_{1}(x)+iCW_{1,x}(x)$, where $C$ is an arbitrary real and $W_1(x)$ is a real-valued and bounded differentiable function. Here we study in detail nonlinear stationary modes that emerge in complex potentials of this type (for brevity, we call them W-dW potentials). First, we assume that the potential is small and employ asymptotic methods to construct a family of nonlinear modes. Our asymptotic procedure stops at the terms of the $\varepsilon^2$ order, where small $\varepsilon$ characterizes amplitude of the potential. We therefore conjecture that no continuous families of authentic nonlinear modes exist in this case, but "pseudo-modes" that satisfy the equation up to $\varepsilon^2$-error can indeed be found in W-dW potentials. Second, we consider the particular case of a W-dW potential well of finite depth and support our hypothesis with qualitative and numerical arguments. Third, we simulate the nonlinear dynamics of found pseudo-modes and observe that, if the amplitude of W-dW potential is small, then the pseudo-modes are robust and display persistent oscillations around a certain position predicted by the asymptotic expansion. Finally, we study the authentic stationary modes which do not form a continuous family, but exist as isolated points. Numerical simulations reveal dynamical instability of these solutions.
Many recent experimental ultrafast spectroscopy studies have hinted at non-adiabatic dynamics indicating the existence of conical intersections, but their direct observation remains a challenge. The rapid change of the energy gap between the electronic states complicated their observation by requiring bandwidths of several electron volts. In this manuscript, we propose to use the combined information of different X-ray pump-probe techniques to identify the conical intersection. We theoretically study the conical intersection in pyrrole using transient X-ray absorption, time-resolved X-ray spontaneous emission, and linear off-resonant Raman spectroscopy to gather evidence of the curve crossing.
Robots performing tasks in warehouses provide the first example of wide-spread adoption of autonomous vehicles in transportation and logistics. The efficiency of these operations, which can vary widely in practice, are a key factor in the success of supply chains. In this work we consider the problem of coordinating a fleet of robots performing picking operations in a warehouse so as to maximize the net profit achieved within a time period while respecting problem- and robot-specific constraints. We formulate the problem as a weighted set packing problem where the elements in consideration are items on the warehouse floor that can be picked up and delivered within specified time windows. We enforce the constraint that robots must not collide, that each item is picked up and delivered by at most one robot, and that the number of robots active at any time does not exceed the total number available. Since the set of routes is exponential in the size of the input, we attack optimization of the resulting integer linear program using column generation, where pricing amounts to solving an elementary resource-constrained shortest-path problem. We propose an efficient optimization scheme that avoids consideration of every increment within the time windows. We also propose a heuristic pricing algorithm that can efficiently solve the pricing subproblem. While this itself is an important problem, the insights gained from solving these problems effectively can lead to new advances in other time-widow constrained vehicle routing problems.
Following the tremendous success of transformer in natural language processing and image understanding tasks, in this paper, we present a novel point cloud representation learning architecture, named Dual Transformer Network (DTNet), which mainly consists of Dual Point Cloud Transformer (DPCT) module. Specifically, by aggregating the well-designed point-wise and channel-wise multi-head self-attention models simultaneously, DPCT module can capture much richer contextual dependencies semantically from the perspective of position and channel. With the DPCT module as a fundamental component, we construct the DTNet for performing point cloud analysis in an end-to-end manner. Extensive quantitative and qualitative experiments on publicly available benchmarks demonstrate the effectiveness of our proposed transformer framework for the tasks of 3D point cloud classification and segmentation, achieving highly competitive performance in comparison with the state-of-the-art approaches.
We give two proofs to an old result of E. Salehi, showing that the Weyl subalgebra $\mathcal{W}$ of $\ell^\infty(\mathbb{Z})$ is a proper subalgebra of $\mathcal{D}$, the algebra of distal functions. We also show that the family $\mathcal{S}^d$ of strictly ergodic functions in $\mathcal{D}$ does not form an algebra and hence in particular does not coincide with $\mathcal{W}$. We then use similar constructions to show that a function which is a multiplier for strict ergodicity, either within $\mathcal{D}$ or in general, is necessarily a constant. An example of a metric, strictly ergodic, distal flow is constructed which admits a non-strictly ergodic $2$-fold minimal self-joining. It then follows that the enveloping group of this flow is not strictly ergodic (as a $T$-flow). Finally we show that the distal, strictly ergodic Heisenberg nil-flow is relatively disjoint over its largest equicontinuous factor from $|\mathcal{W}|$.
We study extensions of the Election Isomorphism problem, focused on the existence of isomorphic subelections. Specifically, we propose the Subelection Isomorphism and the Maximum Common Subelection problems and study their computational complexity and approximability. Using our problems in experiments, we provide some insights into the nature of several statistical models of elections.
The High Altitude Water Cherenkov (HAWC) observatory and the High Energy Stereoscopic System (H.E.S.S.) are two leading instruments in the ground-based very-high-energy gamma-ray domain. HAWC employs the water Cherenkov detection (WCD) technique, while H.E.S.S. is an array of Imaging Atmospheric Cherenkov Telescopes (IACTs). The two facilities therefore differ in multiple aspects, including their observation strategy, the size of their field of view and their angular resolution, leading to different analysis approaches. Until now, it has been unclear if the results of observations by both types of instruments are consistent: several of the recently discovered HAWC sources have been followed up by IACTs, resulting in a confirmed detection only in a minority of cases. With this paper, we go further and try to resolve the tensions between previous results by performing a new analysis of the H.E.S.S. Galactic plane survey data, applying an analysis technique comparable between H.E.S.S. and HAWC. Events above 1 TeV are selected for both datasets, the point spread function of H.E.S.S. is broadened to approach that of HAWC, and a similar background estimation method is used. This is the first detailed comparison of the Galactic plane observed by both instruments. H.E.S.S. can confirm the gamma-ray emission of four HAWC sources among seven previously undetected by IACTs, while the three others have measured fluxes below the sensitivity of the H.E.S.S. dataset. Remaining differences in the overall gamma-ray flux can be explained by the systematic uncertainties. Therefore, we confirm a consistent view of the gamma-ray sky between WCD and IACT techniques.
Cactus networks were introduced by Lam as a generalization of planar electrical networks. He defined a map from these networks to the Grassmannian Gr($n+1,2n$) and showed that the image of this map, $\mathcal X_n$ lies inside the totally nonnegative part of this Grassmannian. In this paper, we show that $\mathcal X_n$ is exactly the elements of Gr($n+1,2n$) that are both totally nonnegative and isotropic for a particular skew-symmetric bilinear form. For certain classes of cactus networks, we also explicitly describe how to turn response matrices and effective resistance matrices into points of Gr($n+1,2n$) given by Lam's map. Finally, we discuss how our work relates to earlier studies of total positivity for Lagrangian Grassmannians.
We propose a new method for accelerating the computation of a concurrency relation, that is all pairs of places in a Petri net that can be marked together. Our approach relies on a state space abstraction, that involves a mix between structural reductions and linear algebra, and a new data-structure that is specifically designed for our task. Our algorithms are implemented in a tool, called Kong, that we test on a large collection of models used during the 2020 edition of the Model Checking Contest. Our experiments show that the approach works well, even when a moderate amount of reductions applies.
Ethics is sometimes considered to be too abstract to be meaningfully implemented in artificial intelligence (AI). In this paper, we reflect on other aspects of computing that were previously considered to be very abstract. Yet, these are now accepted as being done very well by computers. These tasks have ranged from multiple aspects of software engineering to mathematics to conversation in natural language with humans. This was done by automating the simplest possible step and then building on it to perform more complex tasks. We wonder if ethical AI might be similarly achieved and advocate the process of automation as key step in making AI take ethical decisions. The key contribution of this paper is to reflect on how automation was introduced into domains previously considered too abstract for computers.
In this paper, we investigate the decentralized statistical inference problem, where a network of agents cooperatively recover a (structured) vector from private noisy samples without centralized coordination. Existing optimization-based algorithms suffer from issues of model mismatch and poor convergence speed, and thus their performance would be degraded, provided that the number of communication rounds is limited. This motivates us to propose a learning-based framework, which unrolls well-noted decentralized optimization algorithms (e.g., Prox-DGD and PG-EXTRA) into graph neural networks (GNNs). By minimizing the recovery error via end-to-end training, this learning-based framework resolves the model mismatch issue. Our convergence analysis (with PG-EXTRA as the base algorithm) reveals that the learned model parameters may accelerate the convergence and reduce the recovery error to a large extent. The simulation results demonstrate that the proposed GNN-based learning methods prominently outperform several state-of-the-art optimization-based algorithms in convergence speed and recovery error.