Moncef Gabbouj

PhD & Professor

IEEE Fellow, Member of the European Academy of Sciences, Member of the Academy of Finland

University of Tampere, The Republic of Finland

Title：Legacy of the 1943 McCulloch & Pitts Neuron Model

Abstract：Rethinking Deep Learning (DL) by reconsidering the 1943 McCulloch & Pitts core neuron model used in all ANN architectures for various applications. DL outperformed many traditional approaches in numerous fields of science. However, DL comes at a price of high computational cost and follows mostly a Blackbox approach. Targeting Green Learning, we aim to develop more computationally efficient Artificial Neural Networks, called Operational Neural Networks as alternatives to conventional Convolutional Neural Networks (CNNs). ONNs can perform any linear or non-linear transformation with a proper combination of “nodal” and “pool” operators. This is a great leap towards expanding the neuron’s learning capacity in CNNs, which thus far required the use of a single nodal operator for all synaptic connections for each neuron. This restriction has been lifted by introducing a superior neuron called the “generative neuron” where each nodal operator can be customized during the training to maximize learning. As a result, the network can self-organize the nodal operators of its neurons’ connections. Self-Organized ONNs (Self-ONNs) equipped with superior generative neurons can achieve diversity even with a compact configuration. We shall explore the use of neural network models equipped with the generative and the super neuron in several applications.

Yanjun Liu

PhD & Professor

Recipient of the National Outstanding Youth Science Fund, Recipient of the National Excellent Youth Science Fund

Liaoning University of Technology, China

Title: Intelligent Unmanned Systems: Modeling and Safety Control Theory

Abstract: In recent years, unmanned systems have been widely applied across numerous fields, including military defense, industrial manufacturing, public services, and aerospace technology. The seamless integration of intelligent unmanned systems with emerging technologies has not only enhanced national defense security but also driven technological innovation, economic growth, and the cultivation of new productive forces. This integration plays a critical role in promoting the development of strategic emerging industries and shaping the future of the industrial sector. This report will focus on a series of research issues related to modeling and intelligent control of uncertain systems. The discussion will cover control methods for uncertain nonlinear unmanned systems, including state/output feedback control, robust design, and optimization methods. It will further delve into the challenges of safety control for uncertain systems, investigating intelligent adaptive control methods under various constraints, including constant and time-varying constraints, as well as constraint transformation scenarios. Finally, using a physical experimental simulation platform for intelligent safety control of unmanned systems, practical validation and effectiveness of the proposed control methods will be demonstrated.

Dongrui Wu

PhD & Professor

IEEE Fellow, Associate Editor of IEEE Transactions on Fuzzy Systems, National High-Level Talent, Associate Dean of the School of Artificial Intelligence and Automation

Huazhong University of Science and Technology, China

Title：Knowledge-Data Fusion for Brain-Computer Interface Decoding

Abstract：A brain-computer interface (BCI) enables direct communication between the brain and external devices. Accurate decoding of Electroencephalogram (EEG) signals is critical for non-invasive BCIs. Most existing such decoding algorithms are purely data-driven, requiring plenty of calibration data from the new subject, which is time-consuming and user-unfriendly. This talk introduces knowledge-data fusion approaches for EEG signal decoding, which improve the decoding accuracy by incorporating expert knowledge into algorithm design.

Sean McLoone

PhD & Professor

Chartered Engineer, Senior Member of the IEEE, Fellow of the IET, Associate Editor of the Transactions of the Institute of Measurement and Control

Queen's University Belfast, United Kingdom

Title：Forward Selection Component Analysis: From Variable Selection to Dynamic Sampling

Abstract：Forward Selection Component Analysis (FSCA) is an unsupervised technique for selecting a compact set of representative variables, offering a more interpretable alternative to Principal Component Analysis (PCA). Unlike PCA’s abstract components, FSCA identifies actual variables that capture the structure of the data, making it especially useful in applications such as wavelength selection in Optical Emission Spectroscopy and spatial sampling in semiconductor manufacturing wafer metrology. Since optimal variable selection is an NP-hard problem, FSCA uses a greedy algorithm to achieve near-optimal results efficiently. In this talk I will begin by introducing FSCA with a motivating example. Then, building on submodular optimization principles, I will introduce a lazy variant that maintains performance while significantly reducing computation time. The talk will conclude with recent work on adapting FSCA to provide a general dynamic sampling methodology for wafer surface metrology.

Ling Wang

PhD & Professor

Editor of Expert Systems with Applications, Editor of Swarm and Evolutionary Computation, Recipient of the National Outstanding Youth Science Foundation

Tsinghua University, China

Title: Learning Mechanism-Driven Optimization for Smart Manufacturing Scheduling

Abstract: With the advancement of science and technology, manufacturing systems are transitioning toward high-end, intelligent, and green development. Flexible, distributed, and green manufacturing systems are becoming the mainstream modern model to meet diverse and dynamic societal demands. Currently, optimization techniques driven by learning mechanisms demonstrate excellent performance in addressing complex optimization problems. However, there remains a need for systematic and in-depth research at the level of innovation mechanisms. This report focuses on the complexities of smart manufacturing systems, including flexibility, sustainability, distribution, and multi-stage processes. It addresses multi-level scheduling optimization problems involving production, transportation, and assembly. By integrating machine learning and intelligent optimization, the study proposes strategy self-generation mechanisms, algorithm self-adaptation mechanisms, and modality self-organization mechanisms, providing theoretical and methodological support for scheduling optimization in smart manufacturing systems.

Xizhao Wang

PhD & Professor

IEEE Fellow, CAAI Fellow, Editor of Machine Learning and Cybernetics

Shenzhen University, China

Title: Uncertainty Modeling for Mitigating Hallucinations in Large Language Models

Abstract: The rise of generative artificial intelligence and large language models in recent years has sparked a revolution in both academic and industrial AI communities. Since the advent of ChatGPT-4, large language models have flourished, demonstrating emergent capabilities in many specific tasks. However, their application across various domains has also exposed significant shortcomings, such as a lack of human-like knowledge reasoning abilities and issues with hallucinations. This report, by examining the overconfidence in deep model reasoning and corresponding calibration mechanisms, proposes an uncertainty modeling-based approach to formalize and mitigate the hallucination problem.

Lizhong Ding

PhD & Professor

National-level Young Talent, Reviewer for top conferences including NeurIPS, ICML, ICLR

Beijing Institute of Technology, China

Title：A Novel Multi-Table Learning Paradigm: Quantifying and Integrating Complementary Information

Abstract：Multi-table data encompasses diverse entities and attributes with underlying interrelationships. However, existing tabular learning methods struggle to characterize and leverage the intrinsic complementarity among tables. To address this limitation, we propose, for the first time, a multi-table learning paradigm designed to quantify and integrate complementary information across multiple tables. Specifically, we define a metric termed Complementary Strength (CS), which combines correlation, similarity, and informativeness to measure the complementarity between tables. A multi-table learning paradigm is introduced, formalizing multi-table tasks and loss functions. Correspondingly, we propose a multi-table learning network (ATCA-Net) that integrates complementary information across diverse tables via adaptive tabular encoders and cross-table attention mechanisms. Extensive experiments demonstrate that the CS metric effectively quantifies inter-table complementarity, and ATCA-Net successfully leverages it. This work establishes the first theoretical and practical foundation for multi-table learning.

C. L. Philip Chen

PhD & Professor

IEEE Life Fellow  AAAS Fellow, IAPR Fellow, Member of Academia Europaea, Member of the European Academy of Sciences and Arts, Foreign Member of the Russian Academy of Engineering

South China University of Technology, China

Title：Discussion on the Development Trends of Artificial Intelligence and the Applications of AIGC

Abstract：One of the latest advancements in artificial intelligence (AI) is human-machine hybrid enhanced intelligence. In the past, AI mostly operated as standalone systems, but in the future, machines and humans will be "hybridized," operating together with humans in the loop—or outside the loop—learning from each other and enhancing each other's intelligence. The recently emerged AIGC/GPT series of human-computer interactive software robots are an example of such human-machine hybrid enhanced intelligence. These platforms require massive amounts of data, powerful computing capabilities, and high-quality algorithms for support. Currently, the "three computing elements" of AI—algorithm, computing power, and data (computational quantity)—along with application scenarios, have become one of the core capabilities driving the development of the digital economy. To accelerate the construction of the digital economy, effectively stimulate innovation through data, speed up the process of digital industrialization and industrial digitization, foster new technologies, industries, and business models, and support high-quality economic development, the state has planned for cities to adopt layouts of technologies such as artificial intelligence and the Internet of Things to promote the development of new industries. This report first introduces the development trends, evolutionary characteristics, and importance of artificial intelligence, and then discusses the recently popular applications of AIGC.

Tong Zhang

PhD & Professor

Recipient of the National Science Fund for Excellent Young Scientists,  Associate Editor for IEEE Transactions on Affective Computing and IEEE Transactions on Computational Social Systems

South China University of Technology, China

Title：Optimization Methods and Applications of Large Language Models in Vertical Domains

Abstract：This report focuses on discussing how to develop usable, practical, and general-purpose large language models (LLMs) for vertical domains. It explores single-modality LLM techniques across various data modalities, including images, text, and speech, and conducts an in-depth analysis of the characteristics of single modalities and their interrelationships with multimodal data under different subtasks. Based on this, a comprehensive multimodal model is constructed. Building on the above research, multiple model optimization strategies for vertical-domain LLMs are provided through studies on parameter-efficient fine-tuning, knowledge transfer fine-tuning, and adaptation to Chinese language scenarios. Finally, general-purpose optimization techniques for vertical domains are designed, including constructing vertical-domain instruction sets and acquiring structured knowledge, designing single-modality agents within vertical models, and leveraging Mixture of Experts (MoE) to transform the comprehensive multimodal model into a multimodal foundation for vertical-domain models. By accomplishing these tasks in a hierarchical and phased manner, this report demonstrates the application of large models in empowering various domain-specific scenarios.

Ju Fan

PhD & Professor

Recipient of the ICDE 2025 Best Paper Runner-Up Award, the ACM SIGMOD Research Highlight Award, and the ACM China Rising Star Award

Renmin University of China, China

Title：Preliminary Study on a Novel Data Science System Integrating Data and Intelligence

Abstract：The level of intelligence and execution performance of data science systems directly impacts the efficiency of unlocking data value. However, for a long time, data science research has often focused on localized optimizations for isolated issues such as data preparation, visualization, and analytical modeling, lacking a holistic consideration of systematicity, generalizability, and process closure. With the rapid development of artificial intelligence technologies, particularly breakthroughs in reasoning large models, multimodal semantic understanding, and agent mechanisms, data science is now facing new opportunities for a paradigm shift in system design. This report will explore how to promote the deep integration of data and intelligence to address key challenges in the full data science workflow regarding intelligence, automation, and self-adaptation, and share the speaker’s recent research and practical progress in related directions.

Jun Huang

PhD & Professor

Reviewer for international journals including TPAMIand TKDE; Standing Director of Anhui Artificial Intelligence Society

Anhui University of Technology, China

Title：Multimodal and Multiview Learning for Time Series Forecasting and Anomaly Detection

Abstract：Time series forecasting and anomaly detection are fundamental tasks in monitoring and analyzing dynamic systems across diverse domains, including industrial monitoring, financial markets, healthcare, and large-scale software systems. The inherent complexity, non-stationarity, and multi-modal nature of real-world time series data pose significant challenges for accurate prediction and reliable anomaly identification. In this report, we present two novel methodologies addressing these challenges. First, we propose Twin Graph-based Anomaly Detection via Attentive Multi-Modal Learning, a framework that effectively integrates heterogeneous data sources through attentive fusion and dual graph representations to enhance anomaly detection in complex, evolving environments. Second, we introduce the Multi-View Multi-Scale Heterogeneous Cascaded Mamba Model for non-stationary time series forecasting, which leverages a hierarchical Mamba-based architecture to capture long-range dependencies and multi-scale patterns from multiple perspectives, improving forecasting robustness under dynamic conditions. Together, these approaches provide advanced solutions for time series analysis, demonstrating strong potential across a wide range of practical applications.

Yingjie Wang

PhD & Professor

CCF Distinguished Member; Member of the Board of Directors of Shandong Computer Society

Yantai University, China

Title：Multi-Space Crowdsensing Computing

Abstract：Multi-space crowdsensing represents an expansion and upgrade of crowdsensing in the spatial dimension. It is no longer confined to ground-based mobile terminals but extends the sensing entities to multiple physical spaces—aerial, space, maritime, and terrestrial—forming a multi-dimensional, heterogeneous, and collaborativelarge-scale sensing network. With the growing emphasis on satellite constellations, the low-altitude economy, and the marine economy, multi-space crowdsensing computing and related information processing research in aerial-space-maritime domains have attracted significant attention. However, multi-space crowdsensing computing faces three core challenges: difficulty in coordinating multi-source sensing, vulnerability to data privacy leakage, and challenges in controlling sensing quality. To address these issues, the speaker has conducted years of dedicated research. This report will focus on the speaker’s optimization studies in multi-space crowdsensing computing, particularly on task allocation, privacy protection, and quality control, as well as preliminary applications of these research outcomes in aerial-space-maritime information networks.