Speaker |
Xidong Feng |
|---|---|
Affiliation |
DeepMind |
Date |
Friday, 06 December 2024 |
Time |
12:00-13:00 |
Location |
Function Space, UCL Centre for Artificial Intelligence, 1st Floor, 90 High Holborn, London WC1V 6BH |
Link |
https://ucl.zoom.us/j/99748820264 |
Event series |
Jump Trading/ELLIS CSML Seminar Series |
Abstract |
Large Language Models (LLMs) have achieved remarkable success in various natural language processing tasks but remain limited in areas such as long-term planning, self-evaluation, and mitigating hallucinations. This talk explores the intersection of Reinforcement Learning (RL) and LLMs, examining how RL has shaped the development of current LLMs and envisioning its transformative potential in training next-generation models. I will discuss three of my recent works that exemplify this direction: (1) ChessGPT, on how to integrate strategic decision-making with natural language reasoning in chess. (2) AlphaZero-like Tree-Search for LLMs, demonstrating how tree-search algorithms can enhance LLM inference and training. (3) Natural Language Reinforcement Learning, a new language-driven decision-making framework that reframes RL components into their language representations. |
Biography |
Xidong Feng is a research scientist at Google DeepMind. His research spans over Large Language Model, Reinforcement Learning, and Multi-agent Learning. He has published over 10 papers in top AI conferences or journals like NeurIPS, ICML and JMLR. He is going to graduate and obtain his Ph.D. at Computer Science, University College London, advised by Prof. Jun Wang. |