arXiv subjects

Bin Wang

Publications and source records attributed to Bin Wang.

At least 757 records · Page 42Linked to original sources

arXiv2020

Graph Structured Network for Image-Text Matching

Image-text matching has received growing interest since it bridges vision and language. The key challenge lies in how to learn correspondence between image and text. Existing works learn coarse correspondence based on object co-occurrence statistics, while failing to learn fine-grained phrase correspondence. In this paper, we present a novel Graph Structured Matching Network (GSMN) to learn fine-grained correspondence. The GSMN explicitly models object, relation and attribute as a structured phrase, which not only allows to learn correspondence of object, relation and attribute separately, but also benefits to learn fine-grained correspondence of structured phrase. This is achieved by node-level matching and structure-level matching. The node-level matching associates each node with its relevant nodes from another modality, where the node can be object, relation or attribute. The associated nodes then jointly infer fine-grained correspondence by fusing neighborhood associations at structure-level matching. Comprehensive experiments show that GSMN outperforms state-of-the-art methods on benchmarks, with relative Recall@1 improvements of nearly 7% and 2% on Flickr30K and MSCOCO, respectively. Code will be released at: https://github.com/CrossmodalGroup/GSMN.

cs.CV↗

arXiv2020

On nonlinearity in hydrodynamic response to the initial geometry in relativistic heavy-ion collisions

In the context of event-by-event hydrodynamic description, we analyze the implications of two models characterized by distinct initial conditions. The initial energy density of the first model adopts a Gaussian-type distribution, while those of the second one are features by high energy peripheral tubes. We calibrate the initial conditions of both models so that their initial probability distribution of eccentricity are mostly identical. Subsequently, the resultant scaled probability distributions of collective flow and the correlations between flow harmonic and eccentricity coefficients are investigated. Besides, the calculations are carried out for particle correlations regarding the symmetric cumulant, mixed harmonics, and nonlinear response coefficients. Although the resultant two-particle correlations possess similar shapes, numerical calculations indicate a subtle difference between the two models. To be specific, the difference resides in more detailed observables such as the probability distributions of elliptic flow as well as Pearson correlation coefficient regarding higher-order harmonics. We discuss several essential aspects concerning the linearity and nonlinearity between initial eccentricities and final state anisotropies. Further implications are addressed.

nucl-th↗

arXiv2020

Feature-Driven Super-Resolution for Object Detection

Although some convolutional neural networks (CNNs) based super-resolution (SR) algorithms yield good visual performances on single images recently. Most of them focus on perfect perceptual quality but ignore specific needs of subsequent detection task. This paper proposes a simple but powerful feature-driven super-resolution (FDSR) to improve the detection performance of low-resolution (LR) images. First, the proposed method uses feature-domain prior which extracts from an existing detector backbone to guide the HR image reconstruction. Then, with the aligned features, FDSR update SR parameters for better detection performance. Comparing with some state-of-the-art SR algorithms with 4$\times$ scale factor, FDSR outperforms the detection performance mAP on MS COCO validation, VOC2007 databases with good generalization to other detection networks.

cs.CV↗

arXiv2020

TEAM: An Taylor Expansion-Based Method for Generating Adversarial Examples

Although Deep Neural Networks(DNNs) have achieved successful applications in many fields, they are vulnerable to adversarial examples.Adversarial training is one of the most effective methods to improve the robustness of DNNs, and it is generally considered as solving a saddle point problem that minimizes risk and maximizes perturbation.Therefore, powerful adversarial examples can effectively replicate the situation of perturbation maximization to solve the saddle point problem.The method proposed in this paper approximates the output of DNNs in the input neighborhood by using the Taylor expansion, and then optimizes it by using the Lagrange multiplier method to generate adversarial examples. If it is used for adversarial training, the DNNs can be effectively regularized and the defects of the model can be improved.

cs.LG↗

arXiv2020

Particle Swarm Optimisation for Evolving Deep Neural Networks for Image Classification by Evolving and Stacking Transferable Blocks

Deep Convolutional Neural Networks (CNNs) have been widely used in image classification tasks, but the process of designing CNN architectures is very complex, so Neural Architecture Search (NAS), automatically searching for optimal CNN architectures, has attracted more and more research interests. However, the computational cost of NAS is often too high to apply NAS on real-life applications. In this paper, an efficient particle swarm optimisation method named EPSOCNN is proposed to evolve CNN architectures inspired by the idea of transfer learning. EPSOCNN successfully reduces the computation cost by minimising the search space to a single block and utilising a small subset of the training set to evaluate CNNs during evolutionary process. Meanwhile, EPSOCNN also keeps very competitive classification accuracy by stacking the evolved block multiple times to fit the whole dataset. The proposed EPSOCNN algorithm is evaluated on CIFAR-10 dataset and compared with 13 peer competitors comprised of deep CNNs crafted by hand, learned by reinforcement learning methods and evolved by evolutionary computation approaches, which shows very promising results by outperforming all of the peer competitors with regard to the classification accuracy, number of parameters and the computational cost.

cs.NE↗

arXiv2020

New probe of gravity: strongly lensed gravitational wave multi-messenger approach

Strong gravitational lensing by galaxies provides us with a unique opportunity to understand the nature of gravity on galactic and extra-galactic scales. In this paper, we propose a new multimessenger approach using data from both gravitational wave (GW) and the corresponding electromagnetic (EM) counterpart to infer the constraint of the modified gravity (MG) theory denoted by the scale dependent phenomenological parameter. To demonstrate the robustness of this approach, we calculate the time-delay predictions by choosing various values of the phenomenological parameters and then compare them with that from the general relativity (GR). For the third generation ground-based GW observatory, with one typical strongly lensed GW+EM event, and assuming that the dominated error from the stellar velocity dispersions is 5\%, the GW time-delay data can distinguish an 18\% MG effect on a scale of tens of kiloparsecs with a $68\%$ confidence level. Assuming GR and a Singular Isothermal Sphere mass model, there exists a simplified consistency relationship between time-delay and imaging data. This relationship does not require for the velocity dispersion measurement, and hence can avoid major uncertainties. By using this relationship, the multimessenger approach is able to distinguish an $8\%$ MG effect. Our results show that the GW multimessenger approach can play an important role in revealing the nature of gravity on the galactic and extra-galactic scales.

astro-ph.CO↗

arXiv2020

Efficient Sentence Embedding via Semantic Subspace Analysis

A novel sentence embedding method built upon semantic subspace analysis, called semantic subspace sentence embedding (S3E), is proposed in this work. Given the fact that word embeddings can capture semantic relationship while semantically similar words tend to form semantic groups in a high-dimensional embedding space, we develop a sentence representation scheme by analyzing semantic subspaces of its constituent words. Specifically, we construct a sentence model from two aspects. First, we represent words that lie in the same semantic group using the intra-group descriptor. Second, we characterize the interaction between multiple semantic groups with the inter-group descriptor. The proposed S3E method is evaluated on both textual similarity tasks and supervised tasks. Experimental results show that it offers comparable or better performance than the state-of-the-art. The complexity of our S3E method is also much lower than other parameterized models.

cs.CL↗

arXiv2020

The Horizon of the McVittie Black Hole: On the Role of the Cosmic Fluid Modeling

In this paper, we investigate the existence and time evolution of the cosmological and event horizons in a McVittie universe whose expansion is driven by the Redlich-Kwong, (Modified) Berthelot, Dieterici, and Peng-Robinson fluids, respectively. The equations of state of these fluids are rich enough to account for both exotic and regular, as well as ideal and non-ideal matter contents of the universe. We show that the cosmological horizon is expanding, while the event horizon is shrinking along the cosmic time evolution. The former achieves larger size for regular types of matter, contrary to the latter. The strength of interactions within the cosmic fluid are shown to play a more important role in affecting the evolution of the event horizon, rather than of the cosmological horizon in the case of a singularity-free universe. While the cosmological horizon always exists during the time evolution, the event horizon can exist only when a certain relationship between the Hawking-Hayward quasi-local mass and the Hubble function is fulfilled. In this manner, we can study the role played by the large-scale physics (cosmic evolution) on the local scale physics (evolution of a black hole).

gr-qc↗

arXiv2020

Holographic Principle and the Second Law in Stephani Cosmology Revisited

We show in a model-independent way that the inhomogeneous cosmological class II Stephani model fulfills both the the cosmological holographic principle, and that the entropy is increasing with time. By this we mean the result does not depend on any assumption on the time evolution of the scale factor, or on the matter content of the Universe, we also do not need to use the numerical values of the cosmological parameters, which are inferred in the framework of the usual homogeneous Friedmann model. Therefore our analysis is not affected by the tension of the current estimates of the Hubble parameter, and does not rely on any model-dependent assumption of the entropy amount at the present epoch. Our analysis allows us to set an upper limit for the inhomogeneity parameter of the model, an upper bound for the size that this type of universe can reach during the time evolution, a lower bound for the entropy abundance, and an estimate of the present day value of the deceleration parameter.

gr-qc↗

arXiv2020

Joint Extraction of Entities and Relations Based on a Novel Decomposition Strategy

Joint extraction of entities and relations aims to detect entity pairs along with their relations using a single model. Prior work typically solves this task in the extract-then-classify or unified labeling manner. However, these methods either suffer from the redundant entity pairs, or ignore the important inner structure in the process of extracting entities and relations. To address these limitations, in this paper, we first decompose the joint extraction task into two interrelated subtasks, namely HE extraction and TER extraction. The former subtask is to distinguish all head-entities that may be involved with target relations, and the latter is to identify corresponding tail-entities and relations for each extracted head-entity. Next, these two subtasks are further deconstructed into several sequence labeling problems based on our proposed span-based tagging scheme, which are conveniently solved by a hierarchical boundary tagger and a multi-span decoding algorithm. Owing to the reasonable decomposition strategy, our model can fully capture the semantic interdependency between different steps, as well as reduce noise from irrelevant entity pairs. Experimental results show that our method outperforms previous work by 5.2%, 5.9% and 21.5% (F1 score), achieving a new state-of-the-art on three public datasets

cs.CL↗

arXiv2020

Centrality dependence of multiplicity fluctuations from a hydrodynamical approach

As one of the possible signals for the whereabouts of the critical point on the QCD phase diagram, recently, the multiplicity fluctuations in heavy-ion collisions have aroused much attention. It is a crucial observable of the Beam Energy Scan program of the Relativistic Heavy Ion Collider. In this work, we investigate the centrality dependence of the multiplicity fluctuations regarding the recent measurements from STAR Collaboration. By employing a hydrodynamical approach, the present study is dedicated to the noncritical aspects of the phenomenon. To be specific, in addition to the thermal fluctuations, finite volume corrections, and resonance decay at the freeze-out surface, the model is focused on the properties of the hydrodynamic expansion of the system and the event-by-event initial fluctuations. It is understood that the real signal of the critical point can only be obtained after appropriately subtracting the background, the latter is investigated in the present work. Besides the experimental data, our results are also compared to those of the hadronic resonance gas, as well as transport models.

hep-ph↗

arXiv2020

A New Software Framework for Traffic Engineering: Path Cardinality and the Effect of Multipath on Residual Capacity

In this paper, we present a new traffic engineering (TE) software framework to analyze, configure, and optimize (with the aid of a linear programming solver) a network for service provisioning. The developed software tool is based on our new data-driven traffic engineering approach that analyzes a large volume of network configuration data generated given the user input. By analyzing the data, one can then make efficient decisions later when designing a traffic engineering solution. We focus on three well-known traffic engineering objective functions: minimum cost routing (MCR), load balancing (LB), and average delay (AD). With this new tool, one can answer numerous traffic engineering questions. For example, what are the differences among the three objective functions? What is the impact of an objective function on link utilization? How many candidate paths are enough to achieve optimality or near-optimality with respect to a specific objective. This new software tool allows us to conveniently perform various experiments and visualize the results for performance analysis. As case studies, this paper presents examples that answer the questions for two traffic engineering problems: (1) how many paths are required to obtain a solution that is within a few percent from the optimal solution and whether that number is fixed for any network size? (2) how the choice of single-path/multi-path routing affects the load in the network? For the first problem, it turns out that the number of paths needed to achieve optimality increases as the number of links in the network increases.

cs.NI↗

arXiv2020

Hydrodynamic results on multiplicity fluctuations in heavy-ion collisions

Multiplicity fluctuations are one of the most crucial observables in the Beam Energy Scan program of the Relativistic Heavy Ion Collider. It is understood that they can be utilized to probe the whereabouts of the critical point on the phase diagram of the QCD matter. However, a significant portion of these fluctuations is, apart from that related to the QCD phase transition, attributed to the other origins, which we refer to as "noncritical" ones. The present study is dedicated to the noncritical aspects of the multiplicity fluctuations in heavy-ion collisions. In particular, we focus on those of dynamical origin, such as the hydrodynamic expansion of the system and the event-by-event initial fluctuations, in addition to the usual thermal fluctuations, finite volume corrections, and resonance decay at the freeze-out surface. The obtained results are compared to those of the hadronic resonance gas model as well as to the experimental data.

nucl-th↗

arXiv2019

Fractional Dark Matter decay: cosmological imprints and observational constraints

If a fraction $f_{\rm dcdm}$ of the Dark Matter decays into invisible and massless particles (so-called "dark radiation") with the decay rate (or inverse lifetime) $Γ_{\rm dcdm}$, such decay will leave distinctive imprints on cosmological observables. With a full consideration of the Boltzmann hierarchy, we calculate the decay-induced impacts not only on the CMB but also on the redshift distortion and the kinetic Sunyaev-Zel'dovich effect, while providing detailed physical interpretations based on evaluating the evolution of gravitational potential. By using the current cosmological data with a combination of Planck 2015, Baryon Acoustic Oscillation and redshift distortion measurements which can improve the constraints, we update the $1σ$ bound on the fraction of decaying DM from $f_{\rm dcdm}\lesssim5.26\%$ to $f_{\rm dcdm}\lesssim2.73\%$ for the short-lived DM (assuming $Γ_{\rm dcdm}/H_0\gtrsim10^4$). However, no constraints are improved from RSD data ($f_{\rm dcdm}\lesssim0.94\%$) for the long-lived DM (i.e., $Γ_{\rm dcdm}/H_0\lesssim10^4$). We also find the fractional DM decay can only slightly reduce the $H_0$ and $σ_8$ tensions, which is consistent with other previous works. Furthermore, our calculations show that the kSZ effect in future would provide a further constraining power on the decaying DM.

astro-ph.CO↗

arXiv2019

Gravitational Waves in Scalar-Tensor-Vector Gravity Theory

In this paper, we study the properties of gravitational waves in the scalar-tensor-vector gravity theory. The polarizations of the gravitational waves are investigated by analyzing the relative motion of the test particles. It is found that the interaction between the matter and vector field in the theory leads to two additional transverse polarization modes. By making use of the polarization content, the stress-energy pseudo-tensor is calculated by employing the perturbed equation method. Besides, the relaxed field equation for the modified gravity in question is derived by using the Landau-Lifshitz formalism suitable to systems with non-negligible self-gravity.

gr-qc↗

arXiv2019

Spot Evasion Attacks: Adversarial Examples for License Plate Recognition Systems with Convolutional Neural Networks

Recent studies have shown convolution neural networks (CNNs) for image recognition are vulnerable to evasion attacks with carefully manipulated adversarial examples. Previous work primarily focused on how to generate adversarial examples closed to source images, by introducing pixel-level perturbations into the whole or specific part of images. In this paper, we propose an evasion attack on CNN classifiers in the context of License Plate Recognition (LPR), which adds predetermined perturbations to specific regions of license plate images, simulating some sort of naturally formed spots (such as sludge, etc.). Therefore, the problem is modeled as an optimization process searching for optimal perturbation positions, which is different from previous work that consider pixel values as decision variables. Notice that this is a complex nonlinear optimization problem, and we use a genetic-algorithm based approach to obtain optimal perturbation positions. In experiments, we use the proposed algorithm to generate various adversarial examples in the form of rectangle, circle, ellipse and spots cluster. Experimental results show that these adversarial examples are almost ignored by human eyes, but can fool HyperLPR with high attack success rate over 93%. Therefore, we believe that this kind of spot evasion attacks would pose a great threat to current LPR systems, and needs to be investigated further by the security community.

cs.CV↗

arXiv2019

Curvature Invariants and Lower Dimensional Black Hole Horizons

It is known that the event horizon of a black hole can often be identified from the zeroes of some curvature invariants. The situation in lower dimensions has not been thoroughly clarified. In this work we investigate both (2+1)- and (1+1)-dimensional black hole horizons of static, stationary and dynamical black holes, identified with the zeroes of scalar polynomial and Cartan curvature invariants, with the purpose of discriminating the different roles played by the Weyl and Riemann curvature tensors. The situations and applicability of the methods are found to be quite different from that in 4-dimensional spacetime. The suitable Cartan invariants employed for detecting the horizon can be interpreted as a local extremum of the tidal force suggesting that the horizon of a black hole is a genuine special hypersurface within the full manifold, contrary to the usual claim that there is nothing special at the horizon, which is said to be a consequence of the equivalence principle.

gr-qc↗

arXiv2019

A game method for improving the interpretability of convolution neural network

Real artificial intelligence always has been focused on by many machine learning researchers, especially in the area of deep learning. However deep neural network is hard to be understood and explained, and sometimes, even metaphysics. The reason is, we believe that: the network is essentially a perceptual model. Therefore, we believe that in order to complete complex intelligent activities from simple perception, it is necessary to con-struct another interpretable logical network to form accurate and reasonable responses and explanations to external things. Researchers like Bolei Zhou and Quanshi Zhang have found many explanatory rules for deep feature extraction aimed at the feature extraction stage of convolution neural network. However, although researchers like Marco Gori have also made great efforts to improve the interpretability of the fully connected layers of the network, the problem is also very difficult. This paper firstly analyzes its reason. Then a method of constructing logical network based on the fully connected layers and extracting logical relation between input and output of the layers is proposed. The game process between perceptual learning and logical abstract cognitive learning is implemented to improve the interpretable performance of deep learning process and deep learning model. The benefits of our approach are illustrated on benchmark data sets and in real-world experiments.

cs.LG↗