On the afternoon of November 8th at 3:30 pm, Professor Gao Yang from the school invited Associate Professor Wen Ying from Shanghai Jiao Tong University to give an academic lecture titled "Self improvement and Reasoning Enhancement of Large Models Based on Reinforcement Feedback".
Abstrackt:The improvement of the capability of Large Language Models (LLMs) relies on the continuous acquisition of high-quality data and feedback signals. Although a large amount of high-quality data has been utilized during the pre training phase, the key to sustained growth lies in continuously introducing new high-quality data. Due to the high cost of manual data production and the difficulty in meeting demand, it has become crucial to explore methods for self iterative generation and data filtering of large models. This lecture will explore the data reproduction process of large models, including three steps: generation, evaluation, and training. The core challenge is to design efficient algorithms and feedback utilization mechanisms to effectively screen and evaluate data. By applying feedback signals at different levels for reinforcement learning, only the most valuable data will be used for iterative training of the model, and the performance of complex reasoning and decision-making tasks in the inference stage will be enhanced.
Reported by:
Wen Ying, Associate Professor and Doctoral Supervisor at the School of Artificial Intelligence, Shanghai Jiao Tong University. His research interests involve multi-agent learning, reinforcement learning, and the application of game theory in it. He obtained a PhD and a research-oriented master's degree in Computer Science from University College London in 2020 and 2016, respectively. His more than forty research achievements have been published at top international conferences in related fields such as ICML, NeurIPS, ICLR, IJCAI, AAMAS, etc., and have won the CoRL 2020 Best System Paper Award and AAMAS 2021 Blue Sky Track Best Paper Award. He has served as ICML for many consecutive years, PC members or reviewers of internationally renowned conferences/journals such as NeurIPS, IJCAI, AAAI, IROS, ICAPS, Operational Research, etc. In 2021, selected for the Shanghai Youth Science and Technology Talent Sail Plan and selected as a high-level overseas talent in Shanghai.
