Research on Language Model Reasoning and Generalization
Speaker (s):
Yue Zhang
Professor
School of Engineering,
Westlake University
| Date: Time: Venue: | | 23 January 2026, Friday 10:30am – 11:30am School of Computing &
Information Systems 2 (SCIS 2)
Level 4, Seminar Room 4-2
Singapore Management University
90 Stamford Road
Singapore 178903 Please register by 22 January 2026. We look forward to seeing you at this research seminar.  | | | | |
|
About the Talk
Reasoning has become one of the focal points of recent large-model research. Starting from logical reasoning, this talk explores the characteristics of reasoning patterns in large models and their differences from human causal cognition. It shows the gap between the statistical nature of large-model reasoning and causal reasoning. Furthermore, this project investigates the out-of-distribution generalization problem in large models brought about by this gap, which is also one of the challenges toward AGI. In the second half of the talk, it introduces some methods for enhancing the causal nature of reasoning, as well as an approach for automatic detection of large-model generated content by leveraging the differences between statistical and causal properties.
About the Speaker
Professor at Westlake University, Vice Dean of the School of Engineering, ACL Fellow. He received a B.S. degree in Computer Science from Tsinghua University in 2003, and a Ph.D. degree in Computer Science from the University of Oxford in 2009. From 2010 to 2012, he conducted postdoctoral research at the University of Cambridge. His main research areas include natural language processing, language models, and trustworthy artificial intelligence. He authored the Cambridge University Press book Natural Language Processing - A Machine Learning Perspective, and wrote the natural language processing entry for the Oxford Bibliography. He served as Program Committee Chair for CCL 2020, EMNLP 2022, and LMG 2025; as General Chair for NLPCC 2026 and AACL 2026; as Editor-in-Chief of LLM Journal; Action Editor of TACL; and Associate Editor of IEEE/ACM TASLP, IEEE TBD, and ACM TALLIP.
