A5100646812- Natural Language Processing Techniques
- Multimodal Machine Learning Applications
- Topic Modeling
- Target Tracking and Data Fusion in Sensor Networks
- Control Systems and Identification
- Anomaly Detection Techniques and Applications
- Quality Function Deployment in Product Design
- Image and Object Detection Techniques
- Advanced Neural Network Applications
- Image and Signal Denoising Methods
- Industrial Technology and Control Systems
- Adaptive Control of Nonlinear Systems
- Distributed systems and fault tolerance
- Wireless Signal Modulation Classification
- Ferroelectric and Negative Capacitance Devices
- Neural Networks and Applications
- Evaluation and Optimization Models
- Philosophy and History of Science
- Adversarial Robustness in Machine Learning
- User Authentication and Security Systems
- Advanced Image Processing Techniques
- Radar Systems and Signal Processing
- Image Processing and 3D Reconstruction
- Semantic Web and Ontologies
- Advanced Malware Detection Techniques
Tsinghua University
2002-2024
Chinese University of Hong Kong
2024
Hong Kong University of Science and Technology
2019-2023
University of Hong Kong
2019-2023
National University of Defense Technology
2023
University of Electronic Science and Technology of China
2020
University Town of Shenzhen
2017-2020
Renmin University of China
2009-2018
University of Science and Technology Beijing
2010-2014
Nanyang Technological University
2010-2013
Deep learning is usually performed in GPU clusters where each worker machine iteratively refines the model parameters by communicating update with Parameter Server (PS). More often than not, workers communicate a synchronous manner, so as to avoid using out-of-dated and make high-quality refinement iteration. However, all synchronize PS simultaneously, communication becomes severe bottleneck. To address this problem, paper we propose Round-Robin Synchronous Parallel (R <sup...
With the development of recurrent neural networks (RNN), various natural language generation (NLG) tasks have boomed in past few years, such as response conversation and poetry generation. However, automatic news comments is anew, challenging not well-studied task NLG. Different from other NLG tasks, this requires contextual relevance between news. In addition, we need to generate diversified comments, because different people usually opinions on same real world. paper, propose a gated...
China has been on a new journey pursuing high-quality development, an important element of which is global value chain (GVC) upgrading. The smart city pilot policy, aims at reshaping the urban development model in China, carried out and regarded conducive to achieving development. Nevertheless, regarding whether how cities could promote GVC upgrading, research scant. We adopted approach time-varying difference-in-differences (DID) used dataset that encompassed 174 prefecture-level between...
Graph Neural Networks (GNNs) are gaining huge traction recently as they achieve state-of-the-art performance on various graph-related problems. GNN training typically follows the standard Message Passing Paradigm, in which SpMM and SDDMM two essential sparse kernels. However, existing GPU kernels inefficient may suffer from load imbalance, dynamics computing, poor memory efficiency, tail effect. We propose new kernels, Hybrid-Parallel (HP-SpMM) (HP-SDDMM), that efficiently perform GPUs with...
Zijie Huang, Daheng Wang, Binxuan Chenwei Zhang, Jingbo Shang, Yan Liang, Zhengyang Xian Li, Christos Faloutsos, Yizhou Sun, Wei Wang. Findings of the Association for Computational Linguistics: ACL 2023.
Successful ITS (Intelligent Tutor System) design depends on a large number of factors. This paper focuses emotional interaction in ITS. An artificial model based HMM(Hidden Markov Model) is proposed. In this model, the teacher's mood and emotion created form probability, as well as, recognition results students' facial expression are regarded actuating signal model. Teacher's affection considered two-layer random signal, which able to establish various character traits by adjusting initial...
Associative memory is essential to realize man-machine cooperation in the natural interaction between humans and robots. The establishment of associative model solve problem. First, based on theory emotional energy, mood spontaneous metastasis stimulate are put forward. Then we can achieve affective computing external excitation combining with Markov chain which about emotions HMM stimulating metastasis. Second, neural network, applied robots forward by calculating state robot's dynamic...
In this paper, we focus on essay generation, which aims at generating an (a paragraph) according to a set of topic words. Automatic generation can be applied many scenarios reduce human workload. Recently the recurrent neural networks (RNN) based methods are proposed solve task. However, RNN-based suffer from incoherence problem and duplication problem. To overcome these shortcomings, propose self-attention retrieval enhanced network for generation. We retrieve sentences relevant words...
Paraphrase generation aims to rewrite a text with different words while keeping the same meaning. Previous work performs task based solely on given dataset ignoring availability of external linguistic knowledge. However, it is intuitive that model can generate more expressive and diverse paraphrase help such To fill this gap, we propose Knowledge-Enhanced Network (KEPN), transformer-based framework leverage knowledge facilitate generation. (1) The integrates synonym information from into...
For the reconstruction of signals based on a sequence observations governed by uncertain dynamic properties, we propose new method using Taylor's formula and tracking differentiators. The can be summarized as follows: Firstly, for with and/or measurement contains noise, use tracking-differentiators to smooth observation results or filter obtain estimation their differential values. Secondly, signal nearby given points, construct Theoretical is provided guarantee precision relevant results....
This emotion based intelligent tutoring system for English instruction is designed to make up the lack of emotional factors in cognitive ability traditional systems. Microsoft Agent used as interactive platform system. Audio and image information methods. Result face recognition one evaluation factors, guide teaching strategies adjustment. During process, it could provide adaptable strategy not only according test score but also current study status students.
Considering that existing line segment detection algorithms may detect a long as several short fragmented segments, novel linking algorithm is proposed in this paper to improve the performance of detection. Since gradient orientations points on Right Linking Segments (RLSs) have better consistency than those Wrong (WLSs), feature descriptor designed for each candidate based orientation information can effectively distinguish RLSs from WLSs. Experiment results testing images show method...
We consider the problem of adaptive filtering and related derivative estimating for signals with uncertain dynamics this paper. Based on control method systems, a new algorithm is proposed. The theoretical analysis special form filter are given. It shown that doesnpsilat depend unmodeled part systemspsila makes no assumptions regarding classification systems. also robust about initial errors observations. And most important all we can get derivatives by as well. will be helpful system...
The similarity degree of hand-written characters in engineering drawings is analyzed. On the basis results, a hierarchical neural network proposed for recognizing drawings. In memorizing and procedure distributed into several subnetworks. Experimental results are also given.
Abstract In this study, we introduced a machine learning method for estimating human walking speed using plantar pressure and acceleration data. A pressure-derivative based with pretest feature selection was proposed to extracted speed-related features from sensors. The maximum, minimum standard deviation of data were also selected as neural network inputs. To improve the generalization ability network, Bayesian regularization adopted. validate performance method, experiments conducted under...