Jinshan Zeng 

Xi'an Jiaotong University, China

BIO: 

Professor at the School of Management of Xi'an Jiaotong University, and a doctoral advisor. He has led six projects, including those funded by the National Natural Science Foundation of China and the Jiangxi Province's 'Double Thousand Plan', as well as the Provincial Exceptional Youth Program. He has published over 80 high-level papers in mainstream journals and conferences in related fields, including 25 papers in IEEE journals and 32 in top-tier journals or conferences recognized by the China Computer Federation and the China Association for Artificial Intelligence. Among these, 3 are ESI highly cited papers, and one paper was ranked fourth in the 'Top Ten Hot Papers in the Field of Mathematics in China', with the highest citation of a single paper exceeding 1500 times (Google Scholar Citation). He holds over 20 authorized invention patents. He is the Recipient of the 2025 International Basic Science Congress Frontier Science Award, the Best Paper Award at the 2018 and 2020 International Congress of Chinese Mathematicians, and the Second Prize of Jiangxi Natural Science Award (as the first author or principal investigator). He has been invited to deliver special 60-minute or invited 45-minute presentations at the International Basic Science Congress and several International Congresses of Chinese Mathematicians, and has served as vice-chairman or forum chairman at over 10 high-level academic conferences worldwide. The main research focus is on the theory and applications of optimization algorithms in artificial intelligence.

Title: Model-Data-Driven Hyperspectral Compressive Snapshot Reconstruction Methods

Abstract: Coded aperture snapshot spectral imaging (CASSI) is an important technique for capturing three-dimensional (3D) hyperspectral images (HSIs), and involves an inverse problem of reconstructing the 3D HSI from its corresponding coded 2D measurements. Existing model-based and learning-based methods either could not explore the implicit feature of different HSIs or require a large amount of paired data for training, resulting in low reconstruction accuracy or poor generalization performance as well as interpretability. In this talk, we introduce three novel model-data-driven HSI reconstruction methods to remedy these deficiencies, through exploiting the global spectral correlation from the HSI itself through a formulation of model-driven low-rank subspace representation and learning the deep priors by some data-driven deep learning schemes. Extensive experiments on several datasets and imaging systems validate the superiority of our methods.

Tinghua Wang

Gannan Normal University, China

BIO: 

Tinghua Wang received the Ph.D. degree in Computer Science from Beijing Jiaotong University, Beijing, China, 2010. From October 2011 to October 2013, he was a Postdoctoral Research Fellow with the Institute of Computer Science and Technology, Peking University, Beijing, China. From March 2016 to March 2017, he was a Visting Scholar with the Centre for Artificial Intelligence, Faculty of Engineering and Information Technology, University of Technology Sydney, Ultimo, NSW, Australia. Prof. Wang is currently a full-time professor with the School of Mathematics and Computer Science, Gannan Normal University, Ganzhou, China. His research interests include artificial intelligence and machine learning. He has authored and coauthored more than 60 papers in IEEE Transactions on Fuzzy Systems, Information Fusion, other refereed journals, and conference proceedings.

Title: HSIC Lasso with applications

Abstract: The Hilbert-Schmidt Independence Criterion Least absolute shrinkage and selection operator (HSIC Lasso), first introduced by Yamada et al. in 2014, represents an advanced feature selection method that integrates the nonlinear dependency measurement capabilities of HSIC with the sparsity-inducing regularization of Lasso. In recent years, HSIC Lasso has demonstrated remarkable versatility beyond its original scope of feature selection, emerging as a powerful tool for diverse applications including statistical inference, kernel learning, and deep learning. This talk will briefly summarize the basic concepts and theory of HSIC Lasso, and its optimization, extended models, and task-oriented algorithms with applications. Furthermore, some potential problems and directions deserving future exploration will be discussed.

Tao'tao Wang

Jiangxi Institute of Technology, China

BIO: 

Professor Wang Taotao, born in Yidu, Hubei Province, is a doctoral supervisor who mainly conducts research in the fields of big data technology and educational management. He serves as the Executive Director of the Jiangxi Computer Society and the Director of the China Simulation Society. He has led or participated in over 20 scientific research projects at various levels, including those funded by the Jiangxi Natural Science Foundation. He has published more than 10 papers in core journals or higher-level publications and has received three Jiangxi Provincial Teaching Achievement Awards. He currently serves as a Member of the Party Committee and Vice President of Jiangxi Institute of Technology.

Title: A Multimodal Data Mining-Based Investigation into the Application Potential of Bacillus Species

Abstract: Traditional approaches to evaluating the application potential of microbial resources in agriculture rely heavily on large-scale experimental testing, which are often inefficient and lack precision. Bacillus species, as core microbial germplasm resources for biopesticides, biofertilizers, and feed additives, urgently require accurate evaluation methods for their efficient development. This study focuses on Bacillus and addresses the challenge of assessing its application potential through a closed-loop framework of “multi-omics data mining – knowledge graph construction – potential prediction – experimental validation.” By doing so, it provides a paradigm for the precise development of agricultural microbial resources.

Moving beyond the limitations of traditional experiment-driven approaches, this study integrates multimodal data (genomic and literature-based) to construct a predictive system for evaluating the application potential of Bacillus strains, significantly improving the efficiency of high-value germplasm screening. The established research framework of “data mining – model construction – experimental validation” offers a generalizable methodology for the precise evaluation of agricultural microbial resources, thereby facilitating the innovation and development of green agricultural products such as microbial pesticides and fertilizers.

Leiyue Yao

Jiangxi University of Traditional Chinese Medicine, China

BIO: 

Yao Leiyue, male, born in July 1982, is a member of the Communist Party of China. He holds a doctorate and is a professor, doctoral supervisor, and a young discipline leader cultivation target at the Computer College of Jiangxi University of Traditional Chinese Medicine. He is also a master's supervisor at Nanchang University and Jiangxi Agricultural University. He is a system analyst. His research focuses include artificial intelligence and data mining, computer vision technology, and big data processing and optimization based on mobile internet. He has been responsible for the National Natural Science Foundation project "Research on Small Sample Complex Human Behavior Recognition Technology Based on Action Semantic Combination" (2024-2027) and the doctoral start-up fund project "Key Technologies for Equipment and Process Digital Twin in Intelligent Manufacturing of Chinese Herbal Tablets". He developed the HPA-UNet tongue image segmentation algorithm, and the related research results have been published in top journals of the Chinese Academy of Sciences such as "IEEE Journal of Biomedical and Health Informatics" and "Applied Soft Computing". He holds an invention patent for "Real-time Action Temporal Positioning Method" and multiple software copyrights.

Title: Research on Small-Sample Complex Human Behavior Recognition Technology Based on Action Semantic Combination

Abstract: Complex human behavior recognition is a core research content in the fields of human-computer interaction and intelligent control, with broad application prospects and high research value. This report aims at zero-sample complex human behavior recognition, and conducts research around four aspects: quantification representation of human behaviors and data augmentation, action sequence positioning, semantic layer feature learning of complex human behaviors. It proposes a three-dimensional matrix representation method for human behaviors based on the compact storage concept, a series of behavior sample enhancement strategies based on human physiological structure and human movement characteristics, and a probing single-step action sequence positioning method based on the priority of large-scale time windows. According to the definition that "any complex human behavior is a combination of basic action sequences", the complex human behavior recognition problem is transformed into two main issues: "the extraction of action verbs and their features in long-time videos" and "judgment of semantic similarity of action combinations". A neural network model that can effectively learn semantic features of complex human behaviors and perform accurate recognition even in the zero-sample scenario is designed and constructed.