Policy Gradient Method under
Hadamard Parameterization and
Its Extension
报告专家：魏轲 教授（复旦大学）
报告时间：7月17日（周四） 15:00-16:00
报告地点：数学学院西303
报告摘要：
Reinforcement learning (RL) is a sequential decision problem and has achieved great success in a variety of areas, for example robotics and large language models. In this talk, we will first discuss the policy gradient method based on the Hadamard parameterization, which is indeed a Riemannian gradient method on the unit sphere. The linear convergence of this method has been established. Then, we will present a general family of policy opitmization method with policy convergence guarantee.
专家简介：
魏轲，复旦大学大数据学院教授，博士生导师。2014年获得牛津大学博士学位，之后在香港科技大学（2014-2015）和加州大学戴维斯分校（2015-2017）从事博士后工作。目前主要研究兴趣包括高维信号与数据处理，强化学习算法与理论；研究成果发表于ACHA、SIAM系列、IEEE TIT、MP、JMLR等领域内权威期刊。
邀请人：唐庆粦
Video:
- Policy Gradient Method under Hadamard Parameterization and Its Extension
- 15:00 - 16:00, 2025-07-17