arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1912.05830cs.LGmath.OCstat.ML
收藏

策略优化中可证明高效的探索

Provably Efficient Exploration in Policy Optimization

Northwestern University(西北大学);Princeton University(普林斯顿大学)

Qi Cai, Zhuoran Yang, Chi Jin, Zhaoran Wang

AI总结:

本文提出乐观近端策略优化算法OPPO,首次在策略优化中实现可证明高效的探索,并在线性函数逼近的对抗性MDP中取得次线性遗憾界。

摘要:

尽管基于策略的强化学习(RL)在实践中取得了巨大成功,但它在理论上的理解远不如基于价值的强化学习,尤其是与后者相比。特别地,如何设计一个可证明高效且包含探索的策略优化算法仍然是一个未解难题。为了弥合这一差距,本文提出了近端策略优化算法的一个乐观变体(OPPO),它遵循策略梯度方向的“乐观版本”。本文证明,在具有线性函数逼近、未知转移以及全信息反馈对抗奖励的情景马尔可夫决策过程问题中,OPPO 实现了 $\tilde{O}(\sqrt{d^2 H^3 T})$ 的遗憾界。其中 $d$ 是特征维度,$H$ 是回合长度,$T$ 是总步数。据我们所知,OPPO 是第一个可证明高效且进行探索的策略优化算法。

英文摘要:

While policy-based reinforcement learning (RL) achieves tremendous successes in practice, it is significantly less understood in theory, especially compared with value-based RL. In particular, it remains elusive how to design a provably efficient policy optimization algorithm that incorporates exploration. To bridge such a gap, this paper proposes an Optimistic variant of the Proximal Policy Optimization algorithm (OPPO), which follows an ``optimistic version'' of the policy gradient direction. This paper proves that, in the problem of episodic Markov decision process with linear function approximation, unknown transition, and adversarial reward with full-information feedback, OPPO achieves $\tilde{O}(\sqrt{d^2 H^3 T} )$ regret. Here $d$ is the feature dimension, $H$ is the episode horizon, and $T$ is the total number of steps. To the best of our knowledge, OPPO is the first provably efficient policy optimization algorithm that explores.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑