A5101995916- Topic Modeling
- Natural Language Processing Techniques
- Context-Aware Activity Recognition Systems
- Auction Theory and Applications
- Service-Oriented Architecture and Web Services
- Advanced Graph Neural Networks
- Medical Research and Treatments
- Higher Education and Teaching Methods
- Pediatric health and respiratory diseases
- Complexity and Algorithms in Graphs
- Advanced Graph Theory Research
- Optimization and Search Problems
- Regional Economic and Spatial Analysis
- Complex Network Analysis Techniques
- Advanced Technologies in Various Fields
- Machine Learning and Algorithms
- Blockchain Technology Applications and Security
- Big Data Technologies and Applications
- Education and Vocational Training
- Handwritten Text Recognition Techniques
- Cardiovascular Health and Risk Factors
- Advanced Malware Detection Techniques
- Emotion and Mood Recognition
- Advanced Image and Video Retrieval Techniques
- Global Educational Reforms and Inequalities
First Affiliated Hospital of Xinxiang Medical University
2022-2024
Guangdong Teachers College of Foreign Language and Arts
2023
Guangdong University of Foreign Studies
2023
Beijing University of Posts and Telecommunications
2023
Shanghai Jiao Tong University
2018-2021
Georgia Institute of Technology
2020-2021
Hubei University of Technology
2020
NARI Group (China)
2020
Wuhan University
2019
Dartmouth College
2011-2012
Zhenghui Wang, Yanru Qu, Liheng Chen, Jian Shen, Weinan Zhang, Shaodian Yimei Gao, Gen Gu, Ken Yong Yu. Proceedings of the 2018 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018.
Named Entity Recognition (NER) is one of the first stages in deep language understanding yet current NER models heavily rely on human-annotated data. In this work, to alleviate dependence labeled data, we propose a Local Additivity based Data Augmentation (LADA) method for semi-supervised NER, which create virtual samples by interpolating sequences close each other. Our approach has two variations: Intra-LADA and Inter-LADA, where performs interpolations among tokens within sentence,...
Purpose Most source recording device identification models for Web media forensics are based on a single feature to complete the task and often have disadvantages of long time poor accuracy. The purpose this paper is propose new method end-to-end network multi-feature fusion devices. Design/methodology/approach This proposes an efficient attention mechanism, so as achieve convenient devices forensics. Findings authors conducted sufficient experiments prove effectiveness that they proposed....
Pervasive computing, the new computing paradigm aiming at providing services anywhere anytime, poses great challenges on dynamic service composition. Existing composition methods can hardly meet requirements of characteristic and heterogeneity in pervasive environment. In this paper, we propose a parameter-based model to accurately describe services. Based model, are aggregated two-layer graph according both semantic syntactic information input output parameters. Moreover, design novel...
Diabetes prediction is an important data science application in the social healthcare domain. There exist two main challenges diabetes task: heterogeneity since demographic and metabolic are of different types, insufficiency number cases a single medical center usually limited. To tackle above challenges, we employ gradient boosting decision trees (GBDT) to handle introduce multi-task learning (MTL) solve insufficiency. this end, Task-wise Split Gradient Boosting Trees (TSGB) proposed for...
In this paper, we present the design of QuAChIE, a Question Answering based Chinese Information Extraction system. QuAChIE mainly depends on well-trained question answering model to extract high-quality triples. The group head entity and relation are regarded as given input text context. For training evaluation each in system, build large-scale information extraction dataset using Wikidata Wikipedia pages by distant supervision. advanced models implemented top pre-trained language enormous...
Three decades of research in communication complexity have led to the invention a number techniques lower bound randomized complexity. The majority these involve properties large submatrices (rectangles) truth-table matrix defining problem. only technique that does not quite fit is information complexity, which has been investigated over last decade. Here, we connect one most powerful "rectangular" techniques: recently-introduced smooth corruption (or "smooth rectangle") bound. We show...
Artificial intelligence (AI) is an important technical support for the development of various industries in today's society. Call center, as window to serve customers, has more and greatly depended on AI technology, which trend become only way call center. This paper first analyzed status AI. Then it reviewed process center its current challenges. Furtherly, this put forward a framework based technology meet these In end application scenarios according characteristics so provide experience...
In graphs with rich texts, incorporating textual information structural would benefit constructing expressive graph embeddings. Among various embedding models, random walk (RW)-based is one of the most popular and successful groups. However, it challenged by two issues when applied on texts: (i) sampling efficiency: deriving from training objective RW-based models (e.g., DeepWalk node2vec), we show that are likely to generate large amounts redundant samples due three main drawbacks. (ii)...
Motor function rehabilitation training is to restore the motor of hand injury maximum extent and meet needs patients’ daily behavior. At present, evaluation work have disadvantages such as relying on subjective experience physicians, unable quantitatively assess loss function, single method. Most these methods only focus independent motion range a organ, lack consideration constraint relationship between adjacent fingers, do not build visual model for it. To end this issue, purpose sports...
We study problems of scheduling jobs on related machines so as to minimize the makespan in setting where are strategic agents. In this problem, each job $j$ has a length $l_{j}$ and machine $i$ private speed $t_{i}$. The running time is $t_{i}l_{j}$. seek mechanism that obtains bids then assign payments have incentive report true speeds allocation also envy-free. show 1. A deterministic envy-free, truthful, individually rational, anonymous cannot approximate strictly better than $2-1/m$, $m$...
Abstract Big data has become an important reference and helper tool for enterprise efficiency improvement. Effective mining analysis can be utilized by enterprises to enhance the user experience of products or develop services based on needs analysis. In order explore application analytics, this study focuses big analytics from cloud computing perspective reviewing massive references online data, as well fieldwork visits specific situations. The results reflect that is used in various...
Abstract The rapid development of computer network technology has brought great convenience to all walks life and made the transmission various information faster. A secure reliable environment a critical impact on functioning network. In this paper, we derive analyze detection probability, false alarm probability judgment threshold new algorithm based results random matrix theory feature-fitting superiority algorithm, which can exhibit better performance. framework theory, reduces problem...
Image harmonization is the process of modifying foreground a composite image in order to achieve cohesive visual consistency with background. Existing works viewed as purely task, using masks distinguish between and background images. However, distinguishing based on human intention more intuitive. In this paper, we developed new multimodal named referring harmonization, which distinguishes text prompts perform harmonization. To collect necessary data for supplement widely used dataset...