English

Nearly Optimal Policy Optimization with Stable at Any Time Guarantee

Machine Learning 2022-12-06 v3 Machine Learning

Abstract

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)\tilde{O}(\sqrt{S^2AH^4K}) where SS is the number of states, AA is the number of actions, HH is the horizon, and KK is the number of episodes, and there is a SH\sqrt{SH} gap compared with the information theoretic lower bound Ω~(SAH3K)\tilde{\Omega}(\sqrt{SAH^3K}). 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)\tilde{O}(\sqrt{SAH^3K} + \sqrt{AH^4K}) regret. When S>HS > 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.

Keywords

Cite

@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

R2 v1 2026-06-24T08:25:32.110Z