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 2106.07046cs.LGcs.DSmath.OC
收藏

平均报酬MDP样本复杂度的紧界

Towards Tight Bounds on the Sample Complexity of Average-reward MDPs

Stanford University(斯坦福大学)

Yujia Jin, Aaron Sidford

摘要:

我们证明了新的上界和下界的样本复杂性,找到一个$ε$最优的政策的无限地平线平均回报马尔可夫决策过程(MDP)访问生成模型。当所有策略的概率转移矩阵的混合时间最多为$t_\mathrm{mix}$时,我们提供了一种算法,该算法使用每个状态-动作对的$\widetilde{O}(t_\mathrm{mix} ε^{-3})$(不经意)样本来解决该问题。此外,我们提供了一个下限表明,在最坏的情况下,任何算法计算不经意的样本的线性依赖于$t_\mathrm{mix}$是必要的。我们得到我们的结果,通过建立无限地平线平均回报MDP和折扣MDP可能进一步的效用之间的连接。

英文摘要:

We prove new upper and lower bounds for sample complexity of finding an $ε$-optimal policy of an infinite-horizon average-reward Markov decision process (MDP) given access to a generative model. When the mixing time of the probability transition matrix of all policies is at most $t_\mathrm{mix}$, we provide an algorithm that solves the problem using $\widetilde{O}(t_\mathrm{mix} ε^{-3})$ (oblivious) samples per state-action pair. Further, we provide a lower bound showing that a linear dependence on $t_\mathrm{mix}$ is necessary in the worst case for any algorithm which computes oblivious samples. We obtain our results by establishing connections between infinite-horizon average-reward MDPs and discounted MDPs of possible further utility.

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。

↑