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 2302.03201cs.LGmath.OCmath.STstat.MLstat.TH
收藏

基于CVaR的近极小极大最优风险敏感强化学习

Near-Minimax-Optimal Risk-Sensitive Reinforcement Learning with CVaR

Cornell University(康奈尔大学);Cornell Tech(康奈尔理工学院)

Kaiwen Wang, Nathan Kallus, Wen Sun

AI总结:

本文研究以CVaR为目标的风险敏感强化学习,提出带Bernstein奖励的UCB算法和奖励驱动值迭代方法,在MAB和表格MDP中分别达到近极小极大最优遗憾界。

摘要:

本文研究风险敏感强化学习(RL),重点关注风险容忍度为$τ$的条件风险价值(CVaR)目标。从多臂赌博机(MABs)开始,我们证明极小极大CVaR遗憾率为$Ω(\sqrt{τ^{-1}AK})$,其中$A$为动作数量,$K$为回合数,并且该下界可通过一种带有新颖Bernstein奖励的上置信界算法达到。对于表格马尔可夫决策过程(MDPs)中的在线强化学习,我们证明了极小极大遗憾下界为$Ω(\sqrt{τ^{-1}SAK})$(采用归一化累积奖励),其中$S$为状态数量,并提出一种新颖的奖励驱动值迭代过程。我们证明,在连续性假设下,该算法达到$\widetilde O(\sqrt{τ^{-1}SAK})$的最优遗憾,并且在一般情况下达到$\widetilde O(τ^{-1}\sqrt{SAK})$的近最优遗憾,对于常数$τ$而言这是极小极大最优的。这一结果改进了现有最优上界。通过对奖励进行适当离散化,我们的算法具有计算高效性。

英文摘要:

In this paper, we study risk-sensitive Reinforcement Learning (RL), focusing on the objective of Conditional Value at Risk (CVaR) with risk tolerance $τ$. Starting with multi-arm bandits (MABs), we show the minimax CVaR regret rate is $Ω(\sqrt{τ^{-1}AK})$, where $A$ is the number of actions and $K$ is the number of episodes, and that it is achieved by an Upper Confidence Bound algorithm with a novel Bernstein bonus. For online RL in tabular Markov Decision Processes (MDPs), we show a minimax regret lower bound of $Ω(\sqrt{τ^{-1}SAK})$ (with normalized cumulative rewards), where $S$ is the number of states, and we propose a novel bonus-driven Value Iteration procedure. We show that our algorithm achieves the optimal regret of $\widetilde O(\sqrt{τ^{-1}SAK})$ under a continuity assumption and in general attains a near-optimal regret of $\widetilde O(τ^{-1}\sqrt{SAK})$, which is minimax-optimal for constant $τ$. This improves on the best available bounds. By discretizing rewards appropriately, our algorithms are computationally efficient.

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。

↑