Seminar on "Modeling LLMs as Complex Systems: Recent Advances" by Dr Hao Liao

Date: 28 November 2025, Friday
Time: 4:30 PM
Venue: FYW-3316, City University of Hong Kong
Zoom Meeting ID: 859 8869 4437
Password: 123456
Speaker: Dr Hao Liao, Shenzhen University

Abstract: This report details key technologies for building next-generation LLM complex agents, focusing on three core capabilities: memory, goals, and character. We will present recent our works: 1) a pre-trained context compressor that enables long-term memory by overcoming context length constraints; 2) a hierarchical reward method for large codebases that guides agents in complex tasks like fault localization; and 3) R-CHAR, a meta-cognitive role-playing framework for generating consistent and believable character behaviors. Together, these works provide a foundation for more powerful and autonomous AI agents.
 
Speaker’s Bio: Dr. Hao Liao is a Tenured Associate Professor at Shenzhen University, holding a PhD from the University of Fribourg, Switzerland (2015). His research focuses on information mining, complex systems, information diffusion, and LLMs. He has led multiple NSFC grants and industry projects with companies like Microsoft and Tencent, and holds over 60 patents. He has published works in top journals and conferences, and has received accolades such as the CCF-Tencent Rhino Bird Award and the Guangdong Provincial Young Top-Talent.
 
WEBINAR WEBSITE:
https://www.ee.cityu.edu.hk/~cccn/
https://www.ee.cityu.edu.hk/~cccn/centre-seminars.htm