arXiv subjects

Yu Wang

Publications and source records attributed to Yu Wang.

At least 1,225 records · Page 68Linked to original sources

arXiv2020

Video-based Person Re-Identification using Gated Convolutional Recurrent Neural Networks

Deep neural networks have been successfully applied to solving the video-based person re-identification problem with impressive results reported. The existing networks for person re-id are designed to extract discriminative features that preserve the identity information. Usually, whole video frames are fed into the neural networks and all the regions in a frame are equally treated. This may be a suboptimal choice because many regions, e.g., background regions in the video, are not related to the person. Furthermore, the person of interest may be occluded by another person or something else. These unrelated regions may hinder person re-identification. In this paper, we introduce a novel gating mechanism to deep neural networks. Our gating mechanism will learn which regions are helpful for person re-identification and let these regions pass the gate. The unrelated background regions or occluding regions are filtered out by the gate. In each frame, the color channels and optical flow channels provide quite different information. To better leverage such information, we generate one gate using the color channels and another gate using the optical flow channels. These two gates are combined to provide a more reliable gate with a novel fusion method. Experimental results on two major datasets demonstrate the performance improvements due to the proposed gating mechanism.

cs.CV↗

arXiv2020

Exceptionally large anomalous Hall effect due to anticrossing of spin-split bands in the antiferromagnetic half-Heusler compound TbPtBi

We have investigated magnetotransport properties and the topological electronic structure of the half-Heusler compound TbPtBi. Our experiments reveal an exceptionally large anomalous Hall effect (AHE) in the canted antiferromagnetic state of TbPtBi with the anomalous Hall angle (AHA) reaching ~0.68-0.76, which is a few times larger than the previously reported record in GdPtBi. First-principles electronic structure and the associated anomalous Hall conductivity were computed in order to interpret the experimental results. Our analysis shows that the AHE in TbPtBi does not originate from the Weyl points but that it is driven by the large net Berry curvature produced by the anticrossing of spin-split bands near the Fermi level in TbPtBi.

cond-mat.mes-hall↗

arXiv2020

Control Synthesis from Linear Temporal Logic Specifications using Model-Free Reinforcement Learning

We present a reinforcement learning (RL) framework to synthesize a control policy from a given linear temporal logic (LTL) specification in an unknown stochastic environment that can be modeled as a Markov Decision Process (MDP). Specifically, we learn a policy that maximizes the probability of satisfying the LTL formula without learning the transition probabilities. We introduce a novel rewarding and path-dependent discounting mechanism based on the LTL formula such that (i) an optimal policy maximizing the total discounted reward effectively maximizes the probabilities of satisfying LTL objectives, and (ii) a model-free RL algorithm using these rewards and discount factors is guaranteed to converge to such policy. Finally, we illustrate the applicability of our RL-based synthesis approach on two motion planning case studies.

cs.RO↗

arXiv2020

UniLMv2: Pseudo-Masked Language Models for Unified Language Model Pre-Training

We propose to pre-train a unified language model for both autoencoding and partially autoregressive language modeling tasks using a novel training procedure, referred to as a pseudo-masked language model (PMLM). Given an input text with masked tokens, we rely on conventional masks to learn inter-relations between corrupted tokens and context via autoencoding, and pseudo masks to learn intra-relations between masked spans via partially autoregressive modeling. With well-designed position embeddings and self-attention masks, the context encodings are reused to avoid redundant computation. Moreover, conventional masks used for autoencoding provide global masking information, so that all the position embeddings are accessible in partially autoregressive language modeling. In addition, the two tasks pre-train a unified language model as a bidirectional encoder and a sequence-to-sequence decoder, respectively. Our experiments show that the unified language models pre-trained using PMLM achieve new state-of-the-art results on a wide range of natural language understanding and generation tasks across several widely used benchmarks.

cs.CL↗

arXiv2020

High-energy gamma-photon polarization in nonlinear Breit-Wheeler pair production and gamma-polarimetry

The interaction of an unpolarized electron beam with a counterpropagating ultraintense linearly polarized laser pulse is investigated in the quantum radiation-dominated regime. We employ a semiclassical Monte Carlo method to describe spin-resolved electron dynamics, photon emissions and polarization, and pair production. Abundant high-energy linearly polarized gamma photons are generated intermediately during this interaction via nonlinear Compton scattering, with an average polarization degree of more than 50%, which further interacting with the laser fields produce electron-positron pairs due to nonlinear Breit-Wheeler process. The photon polarization is shown to significantly affect the pair yield by a factor beyond 10%. The considered signature of the photon polarization in the pair's yield can be experimentally identified in a prospective two-stage setup. Moreover, the signature can serve also for the polarimetry of high-energy high-flux gamma photons with a resolution well below 1% with currently achievable laser facilities.

physics.plasm-ph↗

arXiv2020

Emergence of Competing Stripe Phase near the Mott Transition in Ti-doped Bilayer Calcium Ruthenates

We report the nanoscale imaging of Ti-doped bilayer calcium ruthenates during the Mott metal-insulator transition by microwave impedance microscopy. Different from a typical first-order phase transition where coexistence of the two terminal phases takes place, a new metallic stripe phase oriented along the in-plane crystalline axes emerges inside both the G-type antiferromagnetic insulating state and paramagnetic metallic state. The effect of this electronic state can be observed in macroscopic measurements, allowing us to construct a phase diagram that takes into account the energetically competing phases. Our work provides a model approach to correlate the macroscopic properties and mesoscopic phase separation in complex oxide materials.

cond-mat.str-el↗

arXiv2020

Application of Pre-training Models in Named Entity Recognition

Named Entity Recognition (NER) is a fundamental Natural Language Processing (NLP) task to extract entities from unstructured data. The previous methods for NER were based on machine learning or deep learning. Recently, pre-training models have significantly improved performance on multiple NLP tasks. In this paper, firstly, we introduce the architecture and pre-training tasks of four common pre-training models: BERT, ERNIE, ERNIE2.0-tiny, and RoBERTa. Then, we apply these pre-training models to a NER task by fine-tuning, and compare the effects of the different model architecture and pre-training tasks on the NER task. The experiment results showed that RoBERTa achieved state-of-the-art results on the MSRA-2006 dataset.

cs.CL↗

arXiv2020

Approximately Optimal Spatial Design: How Good is it?

The increasing recognition of the association between adverse human health conditions and many environmental substances as well as processes has led to the need to monitor them. An important problem that arises in environmental statistics is the design of the locations of the monitoring stations for those environmental processes of interest. One particular design criterion for monitoring networks that tries to reduce the uncertainty about predictions of unseen processes is called the maximum-entropy design. However, this design criterion involves a hard optimization problem that is computationally intractable for large data sets. Previous work of Wang et al. (2017) examined a probabilistic model that can be implemented efficiently to approximate the underlying optimization problem. In this paper, we attempt to establish statistically sound tools for assessing the quality of the approximations.

stat.AP↗

arXiv2020

The Sylvester Graphical Lasso (SyGlasso)

This paper introduces the Sylvester graphical lasso (SyGlasso) that captures multiway dependencies present in tensor-valued data. The model is based on the Sylvester equation that defines a generative model. The proposed model complements the tensor graphical lasso (Greenewald et al., 2019) that imposes a Kronecker sum model for the inverse covariance matrix by providing an alternative Kronecker sum model that is generative and interpretable. A nodewise regression approach is adopted for estimating the conditional independence relationships among variables. The statistical convergence of the method is established, and empirical studies are provided to demonstrate the recovery of meaningful conditional dependency graphs. We apply the SyGlasso to an electroencephalography (EEG) study to compare the brain connectivity of alcoholic and nonalcoholic subjects. We demonstrate that our model can simultaneously estimate both the brain connectivity and its temporal dependencies.

stat.ML↗

arXiv2020

SQLFlow: A Bridge between SQL and Machine Learning

Industrial AI systems are mostly end-to-end machine learning (ML) workflows. A typical recommendation or business intelligence system includes many online micro-services and offline jobs. We describe SQLFlow for developing such workflows efficiently in SQL. SQL enables developers to write short programs focusing on the purpose (what) and ignoring the procedure (how). Previous database systems extended their SQL dialect to support ML. SQLFlow (https://sqlflow.org/sqlflow ) takes another strategy to work as a bridge over various database systems, including MySQL, Apache Hive, and Alibaba MaxCompute, and ML engines like TensorFlow, XGBoost, and scikit-learn. We extended SQL syntax carefully to make the extension working with various SQL dialects. We implement the extension by inventing a collaborative parsing algorithm. SQLFlow is efficient and expressive to a wide variety of ML techniques -- supervised and unsupervised learning; deep networks and tree models; visual model explanation in addition to training and prediction; data processing and feature extraction in addition to ML. SQLFlow compiles a SQL program into a Kubernetes-native workflow for fault-tolerable execution and on-cloud deployment. Current industrial users include Ant Financial, DiDi, and Alibaba Group.

cs.DB↗

arXiv2020

Communication Lower Bound in Convolution Accelerators

In current convolutional neural network (CNN) accelerators, communication (i.e., memory access) dominates the energy consumption. This work provides comprehensive analysis and methodologies to minimize the communication for CNN accelerators. For the off-chip communication, we derive the theoretical lower bound for any convolutional layer and propose a dataflow to reach the lower bound. This fundamental problem has never been solved by prior studies. The on-chip communication is minimized based on an elaborate workload and storage mapping scheme. We in addition design a communication-optimal CNN accelerator architecture. Evaluations based on the 65nm technology demonstrate that the proposed architecture nearly reaches the theoretical minimum communication in a three-level memory hierarchy and it is computation dominant. The gap between the energy efficiency of our accelerator and the theoretical best value is only 37-87%.

cs.DC↗

arXiv2019

Fuzzy Logic based Autonomous Parking Systems -- Part III: A Fuzzy Decision Tree System

This paper proposes a robust design of Hybrid Fuzzy Controller for speed and steering angle control in an Intelligent Autonomous Parking System (IAPS). The Hybrid Fuzzy Controller consists of a Base Fuzzy Controller (BFC) and a Supervisory Fuzzy Decision Tree Controller (SFDTC). The BFC evolves from previous work on fuzzy logic control for unmanned parking and it ensures that optimal parking trajectory is achieved with minimal computational cost. SFDTC further increases the system robustness when there is noise in the operating environment. The design of SFDTC combines Decision Tree theory and fuzzy inference mechanism. A data training process is also formulated to achieve better control performance. As a result, IAPS equipped with the new Hybrid Fuzzy Controller with Fuzzy Decision Tree (HFC-FDT) demonstrates optimal performance with faster convergence and minimal deviation from optimal parking trajectory. The detailed design of Supervisory Fuzzy Decision Tree Controller is presented in this paper with a MATLAB simulated experiment which concludes the superior performance of the new design.

eess.SY↗

arXiv2019

Fuzzy Logic based Autonomous Parking Systems -- Part I: An Integrated Multi-functional System

This paper presents an intelligent autonomous parking system with multi-functions. The integrated system consists of two sub-systems, namely the Fuzzy-Based Onboard System and control center. Unlike most current auto-parking experiments, this FBOS enables the car to perform both slot detection and parking under two different parking modes. Facilitated by the control center, real-time monitoring is also achieved to reduce chances of error.

eess.SY↗

arXiv2019

PoTrojan: powerful neural-level trojan designs in deep learning models

With the popularity of deep learning (DL), artificial intelligence (AI) has been applied in many areas of human life. Neural network or artificial neural network (NN), the main technique behind DL, has been extensively studied to facilitate computer vision and natural language recognition. However, the more we rely on information technology, the more vulnerable we are. That is, malicious NNs could bring huge threat in the so-called coming AI era. In this paper, for the first time in the literature, we propose a novel approach to design and insert powerful neural-level trojans or PoTrojan in pre-trained NN models. Most of the time, PoTrojans remain inactive, not affecting the normal functions of their host NN models. PoTrojans could only be triggered in very rare conditions. Once activated, however, the PoTrojans could cause the host NN models to malfunction, either falsely predicting or classifying, which is a significant threat to human society of the AI era. We would explain the principles of PoTrojans and the easiness of designing and inserting them in pre-trained deep learning models. PoTrojans doesn't modify the existing architecture or parameters of the pre-trained models, without re-training. Hence, the proposed method is very efficient.

cs.CR↗

arXiv2019

Subtle metastability of the layered magnetic topological insulator MnBi2Te4 from weak interactions

The metastable layered compound MnBi2Te4 is the first experimentally realized intrinsic antiferromagnetic topological insulator, predicted to host the quantum anomalous Hall effect at high temperatures upon exfoliation to atomically thin layers. While its magnetic ordering and topological properties have generated intensive interest, the mechanism behind its metastability and the ideal crystal synthesis conditions have remained elusive. Here, using a combined first-principles-based approach that considers lattice, charge, and spin degrees of freedom, we investigate the metastability of MnBi2Te4 by calculating the Helmholtz free energy for the reaction Bi2Te3 + MnTe -> MnBi2Te4. We identify a narrow temperature range (767 K to 873 K) in which the compound is stable with respect to the competing binary phases and successfully synthesize high-quality MnBi2Te4 single crystals using the Bi-Te flux method within this range. We also predict the various contributions to the total specific heat, which is consistent with our experimental measurements. Our findings indicate that the degrees of freedom responsible for the van der Waals interaction, magnetic coupling, and nontrivial band topology in layered materials not only enable emergent phenomena but also determine thermodynamic stability. This conclusion lays the foundation for future computational material synthesis of novel layered systems.

cond-mat.mtrl-sci↗

arXiv2019

Local electric field induced spin photocurrent in ReS2

Circular polarized photocurrent is observed near the electrodes on a few-layer ReS2sample at room temperature. For both electrodes, the spatial distribution of the circular polarized photocurrent shows a feature of two wings, with one positive and the other negative. We suggest that this phenomenon arises from the inverse spin Hall effect due to local electric field near the electrode. Bias voltage that modulates this field further controls the sign and magnitude of the inverse spin Hall effect photocurrent. Our research shows that electric field near electrodes has a significant impact on spin transmission operation, hence it could be taken into account for manufacturing spintronic devices in future.

physics.app-ph↗

arXiv2019

"Double-tracking" Characteristic of the Spectral Evolution of GRB 131231A: Synchrotron Origin?

The characteristics of the spectral evolution of the prompt emission of gamma-ray bursts (GRBs), which are closely related to the radiation mechanism (synchrotron or photosphere), are still an unsolved subject. Here, by performing the detailed time-resolved spectral fitting of GRB 131231A, which has a very bright and well-defined single pulse, some interesting spectral evolution features have been found. (i) Both the low-energy spectral index $α$ and the peak energy $E_{\rm p}$ exhibit the "flux-tracking" pattern ("double-tracking" characteristics). (ii) The parameter relations, i.e., $F$ (the energy flux)-$α$, $F$-$E_{\rm p}$, and $E_{\rm p}$-$α$, along with the analogous Yonetoku $E_{\rm p}$-$L_{γ,\rm iso}$ relation for the different time-resolved spectra, show strong monotonous (positive) correlations, both in the rising and the decaying phases. (iii) The values of $α$ do not exceed the synchrotron limit ($α$= -2/3) in all slices across the pulse, favoring the synchrotron origin. We argue that the one-zone synchrotron emission model with the emitter streaming away at a large distance from the central engine can explain all of these special spectral evolution characteristics.

astro-ph.HE↗

arXiv2019

Fuzzy Logic based Autonomous Parking Systems -- Part II: A Hybrid Dual Controller System

This paper presents an intelligent autonomous parking system with Hybrid Fuzzy Controllers (HFCs). The system enables intelligent vehicles to perform slot detection, parallel and vertical parking in a completely unmanned environment. The HFC, constituting of a Base Fuzzy Controller (BFC) and a Supervisory Fuzzy Controller (SFC), optimizes the control logic to counteract external disturbances in parking process by implementing additional fuzzy rule base. Customized HFCs are designed for critical steps in parking, namely turning control and posture stabilization. As a result, more accurate and efficient parking is achieved even when there are uncertainties in vehicle length and friction. Simulated experiments are carried out in MATLAB to verify the robustness of new HFCs and to demonstrate the performance improvement compared with the previous Fuzzy based Onboard System (FBOS).

eess.SY↗