Demystifying the Efficiency of Reinforcement Learning: Two Recent Stories
Abstract
Reinforcement learning (RL) has gained significant interest due to its success in practice. However, the theoretical understanding of many popular RL algorithms remains inadequate. In this talk, the speaker presents two vignettes regarding the effectiveness of RL algorithms. The first vignette demonstrates the minimax optimality of a perturbed model-based RL approach under a generative model. The second vignette covers policy optimization in RL and explores the convergence properties of different methods. These results shed light on the efficacy of RL algorithms in complex scenarios.
Bio
Yuxin Chen is an Assistant Professor in the Department of Electrical Engineering at Princeton University. His research interests include high-dimensional statistics, mathematical optimization, and reinforcement learning. He has received several awards, including the Princeton graduate mentoring award and the AFOSR Young Investigator Award.
