arXiv subjects

Sheng Chen

Publications and source records attributed to Sheng Chen.

At least 145 records · Page 8Linked to original sources

arXiv2021

Gaussian Dynamic Convolution for Efficient Single-Image Segmentation

Interactive single-image segmentation is ubiquitous in the scientific and commercial imaging software. In this work, we focus on the single-image segmentation problem only with some seeds such as scribbles. Inspired by the dynamic receptive field in the human being's visual system, we propose the Gaussian dynamic convolution (GDC) to fast and efficiently aggregate the contextual information for neural networks. The core idea is randomly selecting the spatial sampling area according to the Gaussian distribution offsets. Our GDC can be easily used as a module to build lightweight or complex segmentation networks. We adopt the proposed GDC to address the typical single-image segmentation tasks. Furthermore, we also build a Gaussian dynamic pyramid Pooling to show its potential and generality in common semantic segmentation. Experiments demonstrate that the GDC outperforms other existing convolutions on three benchmark segmentation datasets including Pascal-Context, Pascal-VOC 2012, and Cityscapes. Additional experiments are also conducted to illustrate that the GDC can produce richer and more vivid features compared with other convolutions. In general, our GDC is conducive to the convolutional neural networks to form an overall impression of the image.

cs.CV↗

arXiv2021

Fast and Accurate Quantized Camera Scene Detection on Smartphones, Mobile AI 2021 Challenge: Report

Camera scene detection is among the most popular computer vision problem on smartphones. While many custom solutions were developed for this task by phone vendors, none of the designed models were available publicly up until now. To address this problem, we introduce the first Mobile AI challenge, where the target is to develop quantized deep learning-based camera scene classification solutions that can demonstrate a real-time performance on smartphones and IoT platforms. For this, the participants were provided with a large-scale CamSDD dataset consisting of more than 11K images belonging to the 30 most important scene categories. The runtime of all models was evaluated on the popular Apple Bionic A11 platform that can be found in many iOS devices. The proposed solutions are fully compatible with all major mobile AI accelerators and can demonstrate more than 100-200 FPS on the majority of recent smartphone platforms while achieving a top-3 accuracy of more than 98%. A detailed description of all models developed in the challenge is provided in this paper.

eess.IV↗

arXiv2021

MapGo: Model-Assisted Policy Optimization for Goal-Oriented Tasks

In Goal-oriented Reinforcement learning, relabeling the raw goals in past experience to provide agents with hindsight ability is a major solution to the reward sparsity problem. In this paper, to enhance the diversity of relabeled goals, we develop FGI (Foresight Goal Inference), a new relabeling strategy that relabels the goals by looking into the future with a learned dynamics model. Besides, to improve sample efficiency, we propose to use the dynamics model to generate simulated trajectories for policy training. By integrating these two improvements, we introduce the MapGo framework (Model-Assisted Policy Optimization for Goal-oriented tasks). In our experiments, we first show the effectiveness of the FGI strategy compared with the hindsight one, and then show that the MapGo framework achieves higher sample efficiency when compared to model-free baselines on a set of complicated tasks.

cs.AI↗

arXiv2021

Group Feature Learning and Domain Adversarial Neural Network for aMCI Diagnosis System Based on EEG

Medical diagnostic robot systems have been paid more and more attention due to its objectivity and accuracy. The diagnosis of mild cognitive impairment (MCI) is considered an effective means to prevent Alzheimer's disease (AD). Doctors diagnose MCI based on various clinical examinations, which are expensive and the diagnosis results rely on the knowledge of doctors. Therefore, it is necessary to develop a robot diagnostic system to eliminate the influence of human factors and obtain a higher accuracy rate. In this paper, we propose a novel Group Feature Domain Adversarial Neural Network (GF-DANN) for amnestic MCI (aMCI) diagnosis, which involves two important modules. A Group Feature Extraction (GFE) module is proposed to reduce individual differences by learning group-level features through adversarial learning. A Dual Branch Domain Adaptation (DBDA) module is carefully designed to reduce the distribution difference between the source and target domain in a domain adaption way. On three types of data set, GF-DANN achieves the best accuracy compared with classic machine learning and deep learning methods. On the DMS data set, GF-DANN has obtained an accuracy rate of 89.47%, and the sensitivity and specificity are 90% and 89%. In addition, by comparing three EEG data collection paradigms, our results demonstrate that the DMS paradigm has the potential to build an aMCI diagnose robot system.

cs.LG↗

arXiv2020

Towards Playing Full MOBA Games with Deep Reinforcement Learning

MOBA games, e.g., Honor of Kings, League of Legends, and Dota 2, pose grand challenges to AI systems such as multi-agent, enormous state-action space, complex action control, etc. Developing AI for playing MOBA games has raised much attention accordingly. However, existing work falls short in handling the raw game complexity caused by the explosion of agent combinations, i.e., lineups, when expanding the hero pool in case that OpenAI's Dota AI limits the play to a pool of only 17 heroes. As a result, full MOBA games without restrictions are far from being mastered by any existing AI system. In this paper, we propose a MOBA AI learning paradigm that methodologically enables playing full MOBA games with deep reinforcement learning. Specifically, we develop a combination of novel and existing learning techniques, including curriculum self-play learning, policy distillation, off-policy adaption, multi-head value estimation, and Monte-Carlo tree-search, in training and playing a large pool of heroes, meanwhile addressing the scalability issue skillfully. Tested on Honor of Kings, a popular MOBA game, we show how to build superhuman AI agents that can defeat top esports players. The superiority of our AI is demonstrated by the first large-scale performance test of MOBA AI agent in the literature.

cs.AI↗

arXiv2020

Supervised Learning Achieves Human-Level Performance in MOBA Games: A Case Study of Honor of Kings

We present JueWu-SL, the first supervised-learning-based artificial intelligence (AI) program that achieves human-level performance in playing multiplayer online battle arena (MOBA) games. Unlike prior attempts, we integrate the macro-strategy and the micromanagement of MOBA-game-playing into neural networks in a supervised and end-to-end manner. Tested on Honor of Kings, the most popular MOBA at present, our AI performs competitively at the level of High King players in standard 5v5 games.

cs.AI↗

arXiv2020

Matrix-Monotonic Optimization Part I: Single-Variable Optimization

Matrix-monotonic optimization exploits the monotonic nature of positive semi-definite matrices to derive optimal diagonalizable structures for the matrix variables of matrix-variable optimization problems. Based on the optimal structures derived, the associated optimization problems can be substantially simplified and underlying physical insights can also be revealed. In our work, a comprehensive framework of the applications of matrix-monotonic optimization to multiple-input multiple-output (MIMO) transceiver design is provided for a series of specific performance metrics under various linear constraints. This framework consists of two parts, i.e., Part-I for single-variable optimization and Part-II for multi-variable optimization. In this paper, single-variable matrix-monotonic optimization is investigated under various power constraints and various types of channel state information (CSI) condition. Specifically, three cases are investigated: 1) both the transmitter and receiver have imperfect CSI; 2) perfect CSI is available at the receiver but the transmitter has no CSI; 3) perfect CSI is available at the receiver but the channel estimation error at the transmitter is norm-bounded. In all three cases, the matrix-monotonic optimization framework can be used for deriving the optimal structures of the optimal matrix variables.

cs.IT↗

arXiv2020

Matrix-Monotonic Optimization Part II: Multi-Variable Optimization

In contrast to Part I of this treatise [1] that focuses on the optimization problems associated with single matrix variables, in this paper, we investigate the application of the matrix-monotonic optimization framework in the optimization problems associated with multiple matrix variables. It is revealed that matrix-monotonic optimization still works even for multiple matrix-variate based optimization problems, provided that certain conditions are satisfied. Using this framework, the optimal structures of the matrix variables can be derived and the associated multiple matrix-variate optimization problems can be substantially simplified. In this paper, several specific examples are given, which are essentially open problems. Firstly, we investigate multi-user multiple-input multiple-output (MU- MIMO) uplink communications under various power constraints. Using the proposed framework, the optimal structures of the precoding matrices at each user under various power constraints can be derived. Secondly, we considered the optimization of the signal compression matrices at each sensor under various power constraints in distributed sensor networks. Finally, we investigate the transceiver optimization for multi-hop amplify-and-forward (AF) MIMO relaying networks with imperfect channel state information (CSI) under various power constraints. At the end of this paper, several simulation results are given to demonstrate the accuracy of the proposed theoretical results.

cs.IT↗

arXiv2020

Quantum Criticism: A Tagged News Corpus Analysed for Sentiment and Named Entities

In this research, we continuously collect data from the RSS feeds of traditional news sources. We apply several pre-trained implementations of named entity recognition (NER) tools, quantifying the success of each implementation. We also perform sentiment analysis of each news article at the document, paragraph and sentence level, with the goal of creating a corpus of tagged news articles that is made available to the public through a web interface. Finally, we show how the data in this corpus could be used to identify bias in news reporting.

cs.CY↗

arXiv2020

Compressive Sensing Based Massive Access for IoT Relying on Media Modulation Aided Machine Type Communications

A fundamental challenge of the large-scale Internet-of-Things lies in how to support massive machine-type communications (mMTC). This letter proposes a media modulation based mMTC solution for increasing the throughput, where a massive multi-input multi-output based base station (BS) is used for enhancing the detection performance. For such a mMTC scenario, the reliable active device detection and data decoding pose a serious challenge. By leveraging the sparsity of the uplink access signals of mMTC received at the BS, a compressive sensing based massive access solution is proposed for tackling this challenge. Specifically, we propose a block sparsity adaptive matching pursuit algorithm for detecting the active devices, whereby the block-sparsity of the uplink access signals exhibited across the successive time slots and the structured sparsity of media modulated symbols are exploited for enhancing the detection performance. Moreover, a successive interference cancellation based structured subspace pursuit algorithm is conceived for data demodulation of the active devices, whereby the structured sparsity of media modulation based symbols found in each time slot is exploited for improving the detection performance. Finally, our simulation results verify the superiority of the proposed scheme over state-of-the-art solutions.

cs.IT↗

arXiv2020

Collision-induced breakage of agglomerates in homogenous isotropic turbulence laden with adhesive particles

We carry out direct numerical simulation combined with adhesive discrete element calculations to investigate collision-induced breakage of agglomerates in homogeneous isotropic turbulence. Based on extensive simulation runs, an adhesion parameter Adn is constructed to quantify the possibility of occurrence of sticking, rebound, and breakage events. The collision-induced breakage rate is then formulated based on the Smoluchowski equation and a breakage fraction. The breakage fraction, defined as the fraction of collisions that result in breakage, is then analytically estimated by a convolution of the probability distribution of collision velocity and a universal transfer function. It is shown that the breakage rate decreases exponentially as the adhesion parameter Adn increases for doublets and scales as linear functions of the agglomerate size, with the slope controlled by Adn. These results allow one to estimate the breakage rate for early-stage agglomerates of arbitrary size. Moreover, the role of the flow structure on the collision-induced breakage is also examined. Violent collisions and breakages are more likely caused by particles ejected rapidly from strong vortices and happen in straining sheets. Our results extend the findings of shear-induced fragmentation, forming a more complete picture of breakage of agglomerates in turbulent flows.

physics.flu-dyn↗

arXiv2020

Log orthogonal functions: approximation properties and applications

We present two new classes of orthogonal functions, log orthogonal functions (LOFs) and generalized log orthogonal functions (GLOFs), which are constructed by applying a $\log$ mapping to Laguerre polynomials. We develop basic approximation theory for these new orthogonal functions and apply them to solve several typical fractional differential equations whose solutions exhibit weak singularities. Our error analysis and numerical results show that our methods based on the new orthogonal functions are particularly suitable for functions which have weak singularities at one endpoint, and can lead to exponential convergence rate, as opposed to low algebraic rates if usual orthogonal polynomials are used.

math.NA↗

arXiv2020

Hybrid Transceiver Optimization for Multi-Hop Communications

Multi-hop communication with the aid of large-scale antenna arrays will play a vital role in future emergence communication systems. In this paper, we investigate amplify-and-forward based and multiple-input multiple-output assisted multi-hop communication, in which all nodes employ hybrid transceivers. Moreover, channel errors are taken into account in our hybrid transceiver design. Based on the matrix-monotonic optimization framework, the optimal structures of the robust hybrid transceivers are derived. By utilizing these optimal structures, the optimizations of analog transceivers and digital transceivers can be separated without loss of optimality. This fact greatly simplifies the joint optimization of analog and digital transceivers. Since the optimization of analog transceivers under unit-modulus constraints is non-convex, a projection type algorithm is proposed for analog transceiver optimization to overcome this difficulty. Based on the derived analog transceivers, the optimal digital transceivers can then be derived using matrix-monotonic optimization. Numeral results obtained demonstrate the performance advantages of the proposed hybrid transceiver designs over other existing solutions.

cs.IT↗

arXiv2019

Channel Fingerprint Based Beam Tracking for Millimeter Wave Communications

Beamforming structures with fixed beam codebooks provide economical solutions for millimeter wave (mmWave) communications due to the low hardware cost. However, the training overhead to search for the optimal beamforming configuration is proportional to the codebook size. To improve the efficiency of beam tracking, we propose a beam tracking scheme based on the channel fingerprint database, which comprises mappings between statistical beamforming gains and user locations. The scheme tracks user movement by utilizing the trained beam configurations and estimating the gains of beam configurations that are not trained. Simulations show that the proposed scheme achieves significant beamforming performance gains over existing beam tracking schemes.

cs.IT↗

arXiv2019

Majorization-Minimization Aided Hybrid Transceivers for MIMO Interference Channels

The potential of deploying large-scale antenna arrays in future wireless systems has stimulated extensive research on hybrid transceiver designs aiming to approximate the optimal fully-digital schemes with much reduced hardware cost and signal processing complexity. Generally, this hybrid transceiver structure requires a joint design of analog and digital processing to enable both beamsteering and spatial multiplexing gains. In this paper, we develop various weighted mean-square-error minimization (WMMSE) based hybrid transceiver designs over multiple-input multiple-output (MIMO) interference channels at both millimeter wave (mmWave) and microwave frequencies. Firstly, a heuristic joint design of hybrid precoder and combiner using alternating optimization is proposed, in which the majorization-minimization (MM) method is utilized to design the analog precoder and combiner with unit-modulus constraints. It is validated that this scheme achieves the comparable performance to the WMMSE fully-digital solution. To further reduce the complexity, a phase projection-based two-stage scheme is proposed to decouple the designs of analog and digital precoder combiner. Secondly, inspired by the fully-digital solutions based on the block-diagonalization zero-forcing (BD-ZF) and signal-to-leakage-plus-noise ratio (SLNR) criteria, low-complexity MMbased BD-ZF and SLNR hybrid designs are proposed to well approximate the corresponding fully-digital solutions. Thirdly, the partially-connected hybrid structure for reducing system hardware cost and power consumption is considered, for which the MM-based alternating optimization still works. Numerical results demonstrate the similar or superior performance of all the above proposed schemes over the existing benchmarks.

cs.IT↗

arXiv2019

Closed-Loop Sparse Channel Estimation for Wideband Millimeter-Wave Full-Dimensional MIMO Systems

This paper proposes a closed-loop sparse channel estimation (CE) scheme for wideband millimeter-wave hybrid full-dimensional multiple-input multiple-output and time division duplexing based systems, which exploits the channel sparsity in both angle and delay domains. At the downlink CE stage, random transmit precoder is designed at base station (BS) for channel sounding, and receive combiners at user devices (UDs) are designed to visualize hybrid array as a low-dimensional digital array for facilitating the multi-dimensional unitary ESPRIT (MDU-ESPRIT) algorithm to estimate respective angle-of-arrivals (AoAs). At the uplink CE stage, the estimated downlink AoAs, namely, uplink angle-of-departures (AoDs), are exploited to design multi-beam transmit precoder at UDs to enable BS to estimate the uplink AoAs, i.e., the downlink AoDs, and delays of different UDs using the MDU-ESPRIT algorithm based on the designed receive combiners at BS. Furthermore, a maximum likelihood approach is proposed to pair the channel parameters acquired at the two stages, and the path gains are then obtained using least squares estimator. According to spectrum estimation theory, our solution can acquire the super-resolution estimations of the AoAs/AoDs and delays of sparse multipath components with low training overhead. Simulation results verify the better CE performance and lower computational complexity of our solution over existing state-of-the-art approaches.

cs.IT↗

arXiv2019

Integral points on twisted Markoff surfaces

We study the integral Hasse principle for affine varieties of the form ax^2+y^2+z^2-xyz=m ,using Brauer-Manin obstruction, and we produce examples whose Brauer groups include 4-torsion elements .We will construct their explicit representatives and compute local invariants in special places.

math.NT↗

arXiv2019

DeepSquare: Boosting the Learning Power of Deep Convolutional Neural Networks with Elementwise Square Operators

Modern neural network modules which can significantly enhance the learning power usually add too much computational complexity to the original neural networks. In this paper, we pursue very efficient neural network modules which can significantly boost the learning power of deep convolutional neural networks with negligible extra computational cost. We first present both theoretically and experimentally that elementwise square operator has a potential to enhance the learning power of neural networks. Then, we design four types of lightweight modules with elementwise square operators, named as Square-Pooling, Square-Softmin, Square-Excitation, and Square-Encoding. We add our four lightweight modules to Resnet18, Resnet50, and ShuffleNetV2 for better performance in the experiment on ImageNet 2012 dataset. The experimental results show that our modules can bring significant accuracy improvements to the base convolutional neural network models. The performance of our lightweight modules is even comparable to many complicated modules such as bilinear pooling, Squeeze-and-Excitation, and Gather-Excite. Our highly efficient modules are particularly suitable for mobile models. For example, when equipped with a single Square-Pooling module, the top-1 classification accuracy of ShuffleNetV2-0.5x on ImageNet 2012 is absolutely improved by 1.45% with no additional parameters and negligible inference time overhead.

cs.CV↗