Timezone: »
Poster
Provable Self-Play Algorithms for Competitive Reinforcement Learning
Yu Bai · Chi Jin
Wed Jul 15 09:00 AM -- 09:45 AM & Wed Jul 15 10:00 PM -- 10:45 PM (PDT) @
Self-play, where the algorithm learns by playing against itself without requiring any direct supervision, has become the new weapon in modern Reinforcement Learning (RL) for achieving superhuman performance in practice. However, the majority of exisiting theory in reinforcement learning only applies to the setting where the agent plays against a fixed environment; it remains largely open whether self-play algorithms can be provably effective, especially when it is necessary to manage the exploration/exploitation tradeoff. We study self-play in competitive reinforcement learning under the setting of Markov games, a generalization of Markov decision processes to the two-player case. We introduce a self-play algorithm---Value Iteration with Upper/Lower Confidence Bound (VI-ULCB)---and show that it achieves regret $\mathcal{\tilde{O}}(\sqrt{T})$ after playing $T$ steps of the game, where the regret is measured by the agent's performance against a fully adversarial opponent who can exploit the agent's strategy at any step. We also introduce an explore-then-exploit style algorithm, which achieves a slightly worse regret of $\mathcal{\tilde{O}}(T^{2/3})$, but is guaranteed to run in polynomial time even in the worst case. To the best of our knowledge, our work presents the first line of provably sample-efficient self-play algorithms for competitive reinforcement learning.
Author Information
Yu Bai (Salesforce Research)
Chi Jin (Princeton University)
More from the Same Authors
-
2021 : Near-Optimal Offline Reinforcement Learning via Double Variance Reduction »
Ming Yin · Yu Bai · Yu-Xiang Wang -
2021 : Policy Finetuning: Bridging Sample-Efficient Offline and Online Reinforcement Learning »
Tengyang Xie · Nan Jiang · Huan Wang · Caiming Xiong · Yu Bai -
2021 : The Power of Exploiter: Provable Multi-Agent RL in Large State Spaces »
Chi Jin · Qinghua Liu · Tiancheng Yu -
2021 : Bellman Eluder Dimension: New Rich Classes of RL Problems, and Sample-Efficient Algorithms »
Chi Jin · Qinghua Liu · Sobhan Miryoosefi -
2021 : Sample-Efficient Learning of Stackelberg Equilibria in General-Sum Games »
Yu Bai · Chi Jin · Huan Wang · Caiming Xiong -
2023 : Is RLHF More Difficult than Standard RL? »
Chi Jin -
2023 Poster: Efficient displacement convex optimization with particle gradient descent »
Hadi Daneshmand · Jason Lee · Chi Jin -
2022 Poster: A Simple Reward-free Approach to Constrained Reinforcement Learning »
Sobhan Miryoosefi · Chi Jin -
2022 Spotlight: A Simple Reward-free Approach to Constrained Reinforcement Learning »
Sobhan Miryoosefi · Chi Jin -
2022 Poster: The Power of Exploiter: Provable Multi-Agent RL in Large State Spaces »
Chi Jin · Qinghua Liu · Tiancheng Yu -
2022 Poster: Learning Markov Games with Adversarial Opponents: Efficient Algorithms and Fundamental Limits »
Qinghua Liu · Yuanhao Wang · Chi Jin -
2022 Poster: Near-Optimal Learning of Extensive-Form Games with Imperfect Information »
Yu Bai · Chi Jin · Song Mei · Tiancheng Yu -
2022 Spotlight: Near-Optimal Learning of Extensive-Form Games with Imperfect Information »
Yu Bai · Chi Jin · Song Mei · Tiancheng Yu -
2022 Oral: Learning Markov Games with Adversarial Opponents: Efficient Algorithms and Fundamental Limits »
Qinghua Liu · Yuanhao Wang · Chi Jin -
2022 Spotlight: The Power of Exploiter: Provable Multi-Agent RL in Large State Spaces »
Chi Jin · Qinghua Liu · Tiancheng Yu -
2022 Poster: Provable Reinforcement Learning with a Short-Term Memory »
Yonathan Efroni · Chi Jin · Akshay Krishnamurthy · Sobhan Miryoosefi -
2022 Spotlight: Provable Reinforcement Learning with a Short-Term Memory »
Yonathan Efroni · Chi Jin · Akshay Krishnamurthy · Sobhan Miryoosefi -
2021 : Sample-Efficient Learning of Stackelberg Equilibria in General-Sum Games »
Yu Bai · Chi Jin · Huan Wang · Caiming Xiong -
2021 Poster: Near-Optimal Representation Learning for Linear Bandits and Linear RL »
Jiachen Hu · Xiaoyu Chen · Chi Jin · Lihong Li · Liwei Wang -
2021 Poster: A Sharp Analysis of Model-based Reinforcement Learning with Self-Play »
Qinghua Liu · Tiancheng Yu · Yu Bai · Chi Jin -
2021 Poster: Provable Meta-Learning of Linear Representations »
Nilesh Tripuraneni · Chi Jin · Michael Jordan -
2021 Poster: How Important is the Train-Validation Split in Meta-Learning? »
Yu Bai · Minshuo Chen · Pan Zhou · Tuo Zhao · Jason Lee · Sham Kakade · Huan Wang · Caiming Xiong -
2021 Spotlight: Provable Meta-Learning of Linear Representations »
Nilesh Tripuraneni · Chi Jin · Michael Jordan -
2021 Spotlight: A Sharp Analysis of Model-based Reinforcement Learning with Self-Play »
Qinghua Liu · Tiancheng Yu · Yu Bai · Chi Jin -
2021 Spotlight: How Important is the Train-Validation Split in Meta-Learning? »
Yu Bai · Minshuo Chen · Pan Zhou · Tuo Zhao · Jason Lee · Sham Kakade · Huan Wang · Caiming Xiong -
2021 Spotlight: Near-Optimal Representation Learning for Linear Bandits and Linear RL »
Jiachen Hu · Xiaoyu Chen · Chi Jin · Lihong Li · Liwei Wang -
2021 Poster: Don’t Just Blame Over-parametrization for Over-confidence: Theoretical Analysis of Calibration in Binary Classification »
Yu Bai · Song Mei · Huan Wang · Caiming Xiong -
2021 Poster: Exact Gap between Generalization Error and Uniform Convergence in Random Feature Models »
Zitong Yang · Yu Bai · Song Mei -
2021 Poster: Risk Bounds and Rademacher Complexity in Batch Reinforcement Learning »
Yaqi Duan · Chi Jin · Zhiyuan Li -
2021 Spotlight: Risk Bounds and Rademacher Complexity in Batch Reinforcement Learning »
Yaqi Duan · Chi Jin · Zhiyuan Li -
2021 Spotlight: Exact Gap between Generalization Error and Uniform Convergence in Random Feature Models »
Zitong Yang · Yu Bai · Song Mei -
2021 Spotlight: Don’t Just Blame Over-parametrization for Over-confidence: Theoretical Analysis of Calibration in Binary Classification »
Yu Bai · Song Mei · Huan Wang · Caiming Xiong -
2020 Poster: On Gradient Descent Ascent for Nonconvex-Concave Minimax Problems »
Tianyi Lin · Chi Jin · Michael Jordan -
2020 Poster: Reward-Free Exploration for Reinforcement Learning »
Chi Jin · Akshay Krishnamurthy · Max Simchowitz · Tiancheng Yu -
2020 Poster: Learning Adversarial Markov Decision Processes with Bandit Feedback and Unknown Transition »
Chi Jin · Tiancheng Jin · Haipeng Luo · Suvrit Sra · Tiancheng Yu -
2020 Poster: Provably Efficient Exploration in Policy Optimization »
Qi Cai · Zhuoran Yang · Chi Jin · Zhaoran Wang -
2020 Poster: What is Local Optimality in Nonconvex-Nonconcave Minimax Optimization? »
Chi Jin · Praneeth Netrapalli · Michael Jordan