讲座时间: 2026年08月24日(星期一)下午15:00
讲座地点: 犀浦校区3号教学楼X31541报告厅
主讲人: Wei Wang  教授
主持人: 张晓博  副教授
主讲人简介：
Dr. Wei Wang is a currently a Professor in the Data Science and Analytics Thrust, Information Hub, The Hong Kong University of Science and Technology (Guangzhou), China. Before that, he was a Professor in the School of Computer Science and Engineering, The University of New South Wales, Australia. His current research interests include AI for Data and Knowledge Management, Artificial Intelligence, Large Lanugage Modes, and AI for Science. He has published over two hundred research papers, with most of them in premier journals and conferences. He has won multiple Best Paper (or similar) Awards, including SIGCOMM 2022, ICASSP 2023, ICMR 2021. 
内容简介：
Large language models increasingly support consequential decisions and interactive agents, yet their behavior remains difficult to inspect and control: confidence can be miscalibrated, explanations may be post-hoc, and agent reflection often conflates diagnosis, repair, and selection. This talk explores how LLM behavior can be made more explicit and editable without accessing hidden reasoning. We present two frameworks based on external representations. IDEA extracts decision-relevant knowledge into interpretable factors for calibration and verifiable intervention, while ReSKILL treats agent repair as credit assignment over failure hypotheses, skill patches, and retest outcomes, keeping the language model frozen. Together, they establish a broader principle: let LLMs propose structured semantic content, while explicit mechanisms perform estimation, selection, intervention, and feedback-driven updating. The talk concludes with connections to recent work on understanding and intervening in LLM reasoning.