arXiv subjects

Jun Zhang

Publications and source records attributed to Jun Zhang.

At least 883 records · Page 49Linked to original sources

arXiv2020

Improving RNN transducer with normalized jointer network

Recurrent neural transducer (RNN-T) is a promising end-to-end (E2E) model in automatic speech recognition (ASR). It has shown superior performance compared to traditional hybrid ASR systems. However, training RNN-T from scratch is still challenging. We observe a huge gradient variance during RNN-T training and suspect it hurts the performance. In this work, we analyze the cause of the huge gradient variance in RNN-T training and proposed a new \textit{normalized jointer network} to overcome it. We also propose to enhance the RNN-T network with a modified conformer encoder network and transformer-XL predictor networks to achieve the best performance. Experiments are conducted on the open 170-hour AISHELL-1 and industrial-level 30000-hour mandarin speech dataset. On the AISHELL-1 dataset, our RNN-T system gets state-of-the-art results on AISHELL-1's streaming and non-streaming benchmark with CER 6.15\% and 5.37\% respectively. We further compare our RNN-T system with our well trained commercial hybrid system on 30000-hour-industry audio data and get 9\% relative improvement without pre-training or external language model.

eess.AS↗

arXiv2020

Graph Neural Networks for Scalable Radio Resource Management: Architecture Design and Theoretical Analysis

Deep learning has recently emerged as a disruptive technology to solve challenging radio resource management problems in wireless networks. However, the neural network architectures adopted by existing works suffer from poor scalability, generalization, and lack of interpretability. A long-standing approach to improve scalability and generalization is to incorporate the structures of the target task into the neural network architecture. In this paper, we propose to apply graph neural networks (GNNs) to solve large-scale radio resource management problems, supported by effective neural network architecture design and theoretical analysis. Specifically, we first demonstrate that radio resource management problems can be formulated as graph optimization problems that enjoy a universal permutation equivariance property. We then identify a class of neural networks, named \emph{message passing graph neural networks} (MPGNNs). It is demonstrated that they not only satisfy the permutation equivariance property, but also can generalize to large-scale problems while enjoying a high computational efficiency. For interpretablity and theoretical guarantees, we prove the equivalence between MPGNNs and a class of distributed optimization algorithms, which is then used to analyze the performance and generalization of MPGNN-based methods. Extensive simulations, with power control and beamforming as two examples, will demonstrate that the proposed method, trained in an unsupervised manner with unlabeled samples, matches or even outperforms classic optimization-based algorithms without domain-specific knowledge. Remarkably, the proposed method is highly scalable and can solve the beamforming problem in an interference channel with $1000$ transceiver pairs within $6$ milliseconds on a single GPU.

cs.IT↗

arXiv2020

On the Starspot Centroid Estimation and Calibration Technologies for the Super-high Accuracy Star Tracker

A pointing accuracy better than 1"(3sigma) star tracker plays a significant role for the advanced scientific missions. This thesis makes a series of studies on the typical error sources associated with the positioning and calibration processes, such as the centroiding algorithm, detector noise, motion, focal length drift, variation of the center wavelength of stellar spectrum, the model error, etc., some works of which are listed as follows: IWCOG algorithm is studied for the characteristics of both the accuracy and efficiency. Using Cramer-Rao Lower Bound theorem,centified to have an approximate optimum property for signal with Possion noise and the optimum feature in case of Gaussian noise. In order to enhance the dynamic performance, the generallized dynamic compensation formula is derived to compensate out the error from motion. decoupling of three technical indicators including dynamic performance, pointing accuracy and attitude output rate. To solve the problems exhibited in the in-orbit correction procedure, a brilliant idea, known as the CRLB constraint method, is presented to minimize some small error sources. To fulfill this goal, the full CRLB constaint formulas are derived to build a budget to limit the in-flight parameters. Using this method the overall error are obtained, which in turn results in three significant inferences. The attitude determination process is further optimized, by giving different weights to each star according to their magnitudes and the position over different star field, the QUEST algorithm is thus upgraded. The results show that this method can improve the attitude measurement accuracy statistically by 10-20%. The theory and deductions can be directly applied to various types of star trackers, whether it is a super-high accuracy one, or a high dynamic one.

astro-ph.IM↗

arXiv2020

DeFuzz: Deep Learning Guided Directed Fuzzing

Fuzzing is one of the most effective technique to identify potential software vulnerabilities. Most of the fuzzers aim to improve the code coverage, and there is lack of directedness (e.g., fuzz the specified path in a software). In this paper, we proposed a deep learning (DL) guided directed fuzzing for software vulnerability detection, named DeFuzz. DeFuzz includes two main schemes: (1) we employ a pre-trained DL prediction model to identify the potentially vulnerable functions and the locations (i.e., vulnerable addresses). Precisely, we employ Bidirectional-LSTM (BiLSTM) to identify attention words, and the vulnerabilities are associated with these attention words in functions. (2) then we employ directly fuzzing to fuzz the potential vulnerabilities by generating inputs that tend to arrive the predicted locations. To evaluate the effectiveness and practical of the proposed DeFuzz technique, we have conducted experiments on real-world data sets. Experimental results show that our DeFuzz can discover coverage more and faster than AFL. Moreover, DeFuzz exposes 43 more bugs than AFL on real-world applications.

cs.CR↗

arXiv2020

Dynamic fingerprint of fractionalized excitations in single-crystalline Cu$_3$Zn(OH)$_6$FBr

Quantum spin liquid (QSL) represents a new class of condensed matter states characterized by the long-range many-body entanglement of topological orders. The most prominent feature of the elusive QSL state is the existence of fractionalized spin excitations. Subject to the strong quantum fluctuations, the spin-1/2 antiferromagnetic system on a kagome lattice is the promising candidate for hosting a QSL ground state, but the structurally ideal realization is rare. Here, we report Raman scattering on the single crystalline Cu$_3$Zn(OH)$_6$FBr, and confirm that the ideal kagome structure remains down to low temperatures without any lattice distortion by the angle-resolved polarized Raman responses and second-harmonic-generation measurements. Furthermore, at low temperatures the Raman scattering reveals a continuum of the spin excitations in Cu$_3$Zn(OH)$_6$FBr, in contrast to the sharp magnon peak in the ordered kagome antiferromagnet EuCu$_3$(OH)$_6$Cl$_3$. Such magnetic Raman continuum, in particular, the substantial low-energy one-pair spinon excitation serves as strong evidence for fractionalized spin excitations in Cu$_3$Zn(OH)$_6$FBr.

cond-mat.str-el↗

arXiv2020

Communication-Computation Trade-Off in Resource-Constrained Edge Inference

The recent breakthrough in artificial intelligence (AI), especially deep neural networks (DNNs), has affected every branch of science and technology. Particularly, edge AI has been envisioned as a major application scenario to provide DNN-based services at edge devices. This article presents effective methods for edge inference at resource-constrained devices. It focuses on device-edge co-inference, assisted by an edge computing server, and investigates a critical trade-off among the computation cost of the on-device model and the communication cost of forwarding the intermediate feature to the edge server. A three-step framework is proposed for the effective inference: (1) model split point selection to determine the on-device model, (2) communication-aware model compression to reduce the on-device computation and the resulting communication overhead simultaneously, and (3) task-oriented encoding of the intermediate feature to further reduce the communication overhead. Experiments demonstrate that our proposed framework achieves a better trade-off and significantly reduces the inference latency than baseline methods.

cs.LG↗

arXiv2020

A Low-Complexity Algorithmic Framework for Large-Scale IRS-Assisted Wireless Systems

Intelligent reflecting surfaces (IRSs) are revolutionary enablers for next-generation wireless communication networks, with the ability to customize the radio propagation environment. To fully exploit the potential of IRS-assisted wireless systems, reflective elements have to be jointly optimized with conventional communication techniques. However, the resulting optimization problems pose significant algorithmic challenges, mainly due to the large-scale non-convex constraints induced by the passive hardware implementations. In this paper, we propose a low-complexity algorithmic framework incorporating alternating optimization and gradient-based methods for large-scale IRS-assisted wireless systems. The proposed algorithm provably converges to a stationary point of the optimization problem. Extensive simulation results demonstrate that the proposed framework provides significant speedups compared with existing algorithms, while achieving a comparable or better performance.

cs.IT↗

arXiv2020

Partially-Connected Hybrid Beamforming for Spectral Efficiency Maximization via a Weighted MMSE Equivalence

Hybrid beamforming (HBF) is an attractive technology for practical massive multiple-input and multiple-output (MIMO) millimeter wave (mmWave) systems. Compared with the fully-connected HBF architecture, the partially-connected one can further reduce the hardware cost and power consumption. However, the special block diagonal structure of its analog beamforming matrix brings additional design challenges. In this paper, we develop effective HBF algorithms for spectral efficiency maximization (SEM) in mmWave massive MIMO systems with the partially-connected architecture. One main contribution is that we prove the equivalence of the SEM problem and a matrix weighted sum mean square error minimization (WMMSE) problem, which leads to a convenient algorithmic approach to directly tackle the SEM problem. Specifically, we decompose the equivalent WMMSE problem into the hybrid precoding and hybrid combining subproblems, for which both the optimal digital precoder and combiner have closed-form solutions. For the more challenging analog precoder and combiner, we propose an element iteration based algorithm and a manifold optimization based algorithm. Finally, the hybrid precoder and combiner are alternatively updated. The overall HBF algorithms are proved to monotonously increase the spectral efficiency and converge. Furthermore, we also propose modified algorithms with reduced computational complexity and finite-resolution phase shifters. Simulation results demonstrate that the proposed HBF algorithms achieve significant performance gains over conventional algorithms.

cs.IT↗

arXiv2020

Efficient reinforcement learning control for continuum robots based on Inexplicit Prior Knowledge

Compared to rigid robots that are generally studied in reinforcement learning, the physical characteristics of some sophisticated robots such as soft or continuum robots are higher complicated. Moreover, recent reinforcement learning methods are data-inefficient and can not be directly deployed to the robot without simulation. In this paper, we propose an efficient reinforcement learning method based on inexplicit prior knowledge in response to such problems. We first corroborate the method by simulation and employed directly in the real world. By using our method, we can achieve active visual tracking and distance maintenance of a tendon-driven robot which will be critical in minimally invasive procedures. Codes are available at https://github.com/Skylark0924/TendonTrack.

cs.RO↗

arXiv2020

Microscope Based HER2 Scoring System

The overexpression of human epidermal growth factor receptor 2 (HER2) has been established as a therapeutic target in multiple types of cancers, such as breast and gastric cancers. Immunohistochemistry (IHC) is employed as a basic HER2 test to identify the HER2-positive, borderline, and HER2-negative patients. However, the reliability and accuracy of HER2 scoring are affected by many factors, such as pathologists' experience. Recently, artificial intelligence (AI) has been used in various disease diagnosis to improve diagnostic accuracy and reliability, but the interpretation of diagnosis results is still an open problem. In this paper, we propose a real-time HER2 scoring system, which follows the HER2 scoring guidelines to complete the diagnosis, and thus each step is explainable. Unlike the previous scoring systems based on whole-slide imaging, our HER2 scoring system is integrated into an augmented reality (AR) microscope that can feedback AI results to the pathologists while reading the slide. The pathologists can help select informative fields of view (FOVs), avoiding the confounding regions, such as DCIS. Importantly, we illustrate the intermediate results with membrane staining condition and cell classification results, making it possible to evaluate the reliability of the diagnostic results. Also, we support the interactive modification of selecting regions-of-interest, making our system more flexible in clinical practice. The collaboration of AI and pathologists can significantly improve the robustness of our system. We evaluate our system with 285 breast IHC HER2 slides, and the classification accuracy of 95\% shows the effectiveness of our HER2 scoring system.

eess.IV↗

arXiv2020

Deep Denoising Neural Network Assisted Compressive Channel Estimation for mmWave Intelligent Reflecting Surfaces

Integrating large intelligent reflecting surfaces (IRS) into millimeter-wave (mmWave) massive multi-input-multi-ouput (MIMO) has been a promising approach for improved coverage and throughput. Most existing work assumes the ideal channel estimation, which can be challenging due to the high-dimensional cascaded MIMO channels and passive reflecting elements. Therefore, this paper proposes a deep denoising neural network assisted compressive channel estimation for mmWave IRS systems to reduce the training overhead. Specifically, we first introduce a hybrid passive/active IRS architecture, where very few receive chains are employed to estimate the uplink user-to-IRS channels. At the channel training stage, only a small proportion of elements will be successively activated to sound the partial channels. Moreover, the complete channel matrix can be reconstructed from the limited measurements based on compressive sensing, whereby the common sparsity of angular domain mmWave MIMO channels among different subcarriers is leveraged for improved accuracy. Besides, a complex-valued denoising convolution neural network (CV-DnCNN) is further proposed for enhanced performance. Simulation results demonstrate the superiority of the proposed solution over state-of-the-art solutions.

cs.IT↗

arXiv2020

Magnetic Flux of Active Regions Determining the Eruptive Character of Large Solar Flares

We establish the largest eruptive/confined flare database to date and analyze 322 flares of \emph{GOES} class M1.0 and larger that occurred during 2010$-$2019, i.e., almost spanning the entire solar cycle 24. We find that the total unsigned magnetic flux ($Φ$$_{AR}$) of active regions (ARs) is a key parameter in governing the eruptive character of large flares, with the proportion of eruptive flares exhibiting a strong anti-correlation with $Φ$$_{AR}$. This means that an AR containing a large magnetic flux has a lower probability for the large flares it produces to be associated with a coronal mass ejection (CME). This finding is supported by the high positive correlation we obtained between the critical decay index height and $Φ$$_{AR}$, implying that ARs with a larger $Φ$$_{AR}$ have a stronger magnetic confinement. Moreover, the confined flares originating from ARs larger than 1.0$\times$$10^{23}$ Mx have several characteristics in common: stable filament, slipping magnetic reconnection and strongly sheared post-flare loops. Our findings reveal new relations between the magnetic flux of ARs and the occurrence of CMEs in association with large flares. These relations obtained here provide quantitative criteria for forecasting CMEs and adverse space weather, and have also important implications for "superflares" on solar-type stars and stellar CMEs. The link of database is https://doi.org/10.12149/101030.

astro-ph.SR↗

arXiv2020

EPI-based Oriented Relation Networks for Light Field Depth Estimation

Light field cameras record not only the spatial information of observed scenes but also the directions of all incoming light rays. The spatial and angular information implicitly contain geometrical characteristics such as multi-view or epipolar geometry, which can be exploited to improve the performance of depth estimation. An Epipolar Plane Image (EPI), the unique 2D spatial-angular slice of the light field, contains patterns of oriented lines. The slope of these lines is associated with the disparity. Benefiting from this property of EPIs, some representative methods estimate depth maps by analyzing the disparity of each line in EPIs. However, these methods often extract the optimal slope of the lines from EPIs while ignoring the relationship between neighboring pixels, which leads to inaccurate depth map predictions. Based on the observation that an oriented line and its neighboring pixels in an EPI share a similar linear structure, we propose an end-to-end fully convolutional network (FCN) to estimate the depth value of the intersection point on the horizontal and vertical EPIs. Specifically, we present a new feature-extraction module, called Oriented Relation Module (ORM), that constructs the relationship between the line orientations. To facilitate training, we also propose a refocusing-based data augmentation method to obtain different slopes from EPIs of the same scene point. Extensive experiments verify the efficacy of learning relations and show that our approach is competitive to other state-of-the-art methods. The code and the trained models are available at https://github.com/lkyahpu/EPI_ORM.git.

cs.CV↗

arXiv2020

Do Not Disturb Me: Person Re-identification Under the Interference of Other Pedestrians

In the conventional person Re-ID setting, it is widely assumed that cropped person images are for each individual. However, in a crowded scene, off-shelf-detectors may generate bounding boxes involving multiple people, where the large proportion of background pedestrians or human occlusion exists. The representation extracted from such cropped images, which contain both the target and the interference pedestrians, might include distractive information. This will lead to wrong retrieval results. To address this problem, this paper presents a novel deep network termed Pedestrian-Interference Suppression Network (PISNet). PISNet leverages a Query-Guided Attention Block (QGAB) to enhance the feature of the target in the gallery, under the guidance of the query. Furthermore, the involving Guidance Reversed Attention Module and the Multi-Person Separation Loss promote QGAB to suppress the interference of other pedestrians. Our method is evaluated on two new pedestrian-interference datasets and the results show that the proposed method performs favorably against existing Re-ID methods.

cs.CV↗

arXiv2020

TextureWGAN: Texture Preserving WGAN with MLE Regularizer for Inverse Problems

Many algorithms and methods have been proposed for inverse problems particularly with the recent surge of interest in machine learning and deep learning methods. Among all proposed methods, the most popular and effective method is the convolutional neural network (CNN) with mean square error (MSE). This method has been proven effective in super-resolution, image de-noising, and image reconstruction. However, this method is known to over-smooth images due to the nature of MSE. MSE based methods minimize Euclidean distance for all pixels between a baseline image and a generated image by CNN and ignore the spatial information of the pixels such as image texture. In this paper, we proposed a new method based on Wasserstein GAN (WGAN) for inverse problems. We showed that the WGAN-based method was effective to preserve image texture. It also used a maximum likelihood estimation (MLE) regularizer to preserve pixel fidelity. Maintaining image texture and pixel fidelity is the most important requirement for medical imaging. We used Peak Signal to Noise Ratio (PSNR) and Structure Similarity (SSIM) to evaluate the proposed method quantitatively. We also conducted first-order and second-order statistical image texture analysis to assess image texture.

eess.IV↗

arXiv2020

Metallic Microswimmers Driven up the Wall by Gravity

As a natural and functional behavior, various microorganisms exhibit gravitaxis by orienting and swimming upwards against gravity. Swimming autophoretic nanomotors described herein, comprising bimetallic nanorods, preferentially orient upwards and swim up along a wall, when tail-heavy (i.e. when the density of one of the metals is larger than the other). Through experiment and theory, two mechanisms were identified that contribute to this gravitactic behavior. First, a buoyancy or gravitational torque acts on these rods to align them upwards. Second, hydrodynamic interactions of the rod with the inclined wall induce a fore-aft drag asymmetry on the rods that reinforces their orientation bias and promotes their upward motion.

cond-mat.soft↗

arXiv2020

Flow Field Reconstructions with GANs based on Radial Basis Functions

Nonlinear sparse data regression and generation have been a long-term challenge, to cite the flow field reconstruction as a typical example. The huge computational cost of computational fluid dynamics (CFD) makes it much expensive for large scale CFD data producing, which is the reason why we need some cheaper ways to do this, of which the traditional reduced order models (ROMs) were promising but they couldn't generate a large number of full domain flow field data (FFD) to realize high-precision flow field reconstructions. Motivated by the problems of existing approaches and inspired by the success of the generative adversarial networks (GANs) in the field of computer vision, we prove an optimal discriminator theorem that the optimal discriminator of a GAN is a radial basis function neural network (RBFNN) while dealing with nonlinear sparse FFD regression and generation. Based on this theorem, two radial basis function-based GANs (RBF-GAN and RBFC-GAN), for regression and generation purposes, are proposed. Three different datasets are applied to verify the feasibility of our models. The results show that the performance of the RBF-GAN and the RBFC-GAN are better than that of GANs/cGANs by means of both the mean square error (MSE) and the mean square percentage error (MSPE). Besides, compared with GANs/cGANs, the stability of the RBF-GAN and the RBFC-GAN improve by 34.62% and 72.31%, respectively. Consequently, our proposed models can be used to generate full domain FFD from limited and sparse datasets, to meet the requirement of high-precision flow field reconstructions.

eess.SP↗

arXiv2020

Data-driven Computational Social Science: A Survey

Social science concerns issues on individuals, relationships, and the whole society. The complexity of research topics in social science makes it the amalgamation of multiple disciplines, such as economics, political science, and sociology, etc. For centuries, scientists have conducted many studies to understand the mechanisms of the society. However, due to the limitations of traditional research methods, there exist many critical social issues to be explored. To solve those issues, computational social science emerges due to the rapid advancements of computation technologies and the profound studies on social science. With the aids of the advanced research techniques, various kinds of data from diverse areas can be acquired nowadays, and they can help us look into social problems with a new eye. As a result, utilizing various data to reveal issues derived from computational social science area has attracted more and more attentions. In this paper, to the best of our knowledge, we present a survey on data-driven computational social science for the first time which primarily focuses on reviewing application domains involving human dynamics. The state-of-the-art research on human dynamics is reviewed from three aspects: individuals, relationships, and collectives. Specifically, the research methodologies used to address research challenges in aforementioned application domains are summarized. In addition, some important open challenges with respect to both emerging research topics and research methods are discussed.

cs.SI↗