一、报告时间

2025年9月26日（星期五）15:30

二、报告形式

经管楼101报告厅

三、报告人

吴贤毅 华东师范大学统计学教授

四、报告主题

An introduction to Reinforcement Learning and Empirical Gittins index strategies with ε-explorations

五、报告摘要

This talk consists of two parts: The first is an concise introduction to Reinforcement Learning. The other is on a Gittins index strategies in a special RL models: Markovian MAB model: For Markovian-reward MABs, the optimal policy pulls the arm with the highest Gittins index. When distributions are unknown, an empirical ε-Gittins index policy is proposed that combines ε-exploration with empirical Gittins indices computed via the Largest-Remaining-Index algorithm on estimated models. The theoretical results include the convergence of the empirical indices to true indices; expected discounted returns of the empirical ε-Gittins policy to those of the oracle Gittins policy.

六、报告人简介

吴贤毅，华东师范大学统计学院教授、博士生导师，研究及教学内容涉及统计学、机器学习/人工智能、非寿险精算学、随机调度等领域，在国际主流学术杂志发表过学术论文80余篇，在国内外出版社出版过专著两部，教材一部，内容包括变量选择、多重比较、强化学习、随机准备金评估以及老年护理保险等内容。