arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2302.01248stat.MLcs.LG
收藏

无需模型估计的鲁棒马尔可夫决策过程

Robust Markov Decision Processes without Model Estimation

Peking University(北京大学);University of Alberta(阿尔伯塔大学);OMRON SINIC X(欧姆龙SINIC X)

Wenhao Yang, Han Wang, Tadashi Kozuno, Scott M. Jordan, Zhihua Zhang

AI总结:

本文提出一种无需预言机和模型估计的无模型算法,通过将鲁棒MDP转化为可用随机梯度求解的替代形式,大幅降低存储需求并经实验验证了其高效性。

摘要:

鲁棒马尔可夫决策过程(MDPs)在学习对环境变化不太敏感的鲁棒策略方面正受到越来越多的关注。分析鲁棒MDPs样本效率的研究日益增多。然而,在实践中应用鲁棒MDPs存在两大主要障碍。首先,大多数工作在基于模型的设定下研究鲁棒MDPs,其中需要估计转移概率并需要大量内存$\mathcal{O}(|\mathcal{S}|^2|\mathcal{A}|)$。其次,先前的工作通常假设存在一个强大的预言机,以获取最优解作为求解鲁棒MDPs的中间步骤。然而,在实践中,这种预言机通常并不存在。为了消除预言机,我们将原始的鲁棒MDPs转换为一种替代形式,这使我们能够使用随机梯度方法来求解鲁棒MDPs。此外,我们证明了该替代形式仍然发挥着与原始形式相似的作用。借助这种新的公式化表示,我们设计了一种样本高效的算法,在无模型设定下求解鲁棒MDPs,该算法不需要预言机,并以较低的存储需求$\mathcal{O}(|\mathcal{S}||\mathcal{A}|)$换取从生成模型或马尔可夫链生成样本的能力。最后,我们通过数值实验验证了理论发现,展示了鲁棒MDPs替代形式的效率。

英文摘要:

Robust Markov Decision Processes (MDPs) are receiving much attention in learning a robust policy which is less sensitive to environment changes. There are an increasing number of works analyzing sample-efficiency of robust MDPs. However, there are two major barriers to applying robust MDPs in practice. First, most works study robust MDPs in a model-based regime, where the transition probability needs to be estimated and requires a large amount of memories $\mathcal{O}(|\mathcal{S}|^2|\mathcal{A}|)$. Second, prior work typically assumes a strong oracle to obtain the optimal solution as an intermediate step to solve robust MDPs. However, in practice, such an oracle does not exist usually. To remove the oracle, we transform the original robust MDPs into an alternative form, which allows us to use stochastic gradient methods to solve the robust MDPs. Moreover, we prove the alternative form still plays a similar role as the original form. With this new formulation, we devise a sample-efficient algorithm to solve the robust MDPs in a model-free regime, which does not require an oracle and trades off a lower storage requirement $\mathcal{O}(|\mathcal{S}||\mathcal{A}|)$ with being able to generate samples from a generative model or Markovian chain. Finally, we validate our theoretical findings via numerical experiments, showing the efficiency with the alternative form of robust MDPs.

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。

↑