Policy optimization methods are one of the most widely used classes of Reinforcement Learning (RL) algorithms. However, theoretical understanding of these methods remains insufficient. Even in the episodic (time-inhomogeneous) tabular setting, the state-of-the-art theoretical result of policy-based method in \citet{shani2020optimistic} is only O~(S2AH4K) where S is the number of states, A is the number of actions, H is the horizon, and K is the number of episodes, and there is a SH gap compared with the information theoretic lower bound Ω~(SAH3K). To bridge such a gap, we propose a novel algorithm Reference-based Policy Optimization with Stable at Any Time guarantee (\algnameacro), which features the property "Stable at Any Time". We prove that our algorithm achieves O~(SAH3K+AH4K) regret. When S>H, our algorithm is minimax optimal when ignoring logarithmic factors. To our best knowledge, RPO-SAT is the first computationally efficient, nearly minimax optimal policy-based algorithm for tabular RL.
@article{arxiv.2112.10935,
title = {Nearly Optimal Policy Optimization with Stable at Any Time Guarantee},
author = {Tianhao Wu and Yunchang Yang and Han Zhong and Liwei Wang and Simon S. Du and Jiantao Jiao},
journal= {arXiv preprint arXiv:2112.10935},
year = {2022}
}
Comments
arXiv admin note: text overlap with arXiv:2002.08243 by other authors