arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2403.03361stat.MLcs.LG
收藏

线性老虎机问题的链式信息论界与紧遗憾率

Chained Information-Theoretic bounds and Tight Regret Rate for Linear Bandit Problems

KTH Royal Institute of Technology(皇家理工学院)

Amaury Gouverneur, Borja Rodríguez-Gálvez, Tobias J. Oechtering, Mikael Skoglund

AI总结:

该研究针对线性老虎机问题,依托已有信息论框架与率失真分析,通过链式论证为带度量动作空间的汤普森采样变体建立依赖动作空间度量熵的新界,得到d维场景下$O(d\sqrt{T})$的紧遗憾率。

摘要:

本文研究老虎机问题的一种Thompson Sampling(汤普森采样)算法变体的贝叶斯遗憾。该研究以[Russo和Van Roy, 2015]的信息论框架为基础,更具体地依托[Dong和Van Roy, 2020]的率失真分析——后者在d维线性老虎机场景下证明了遗憾率为$O(d\sqrt{T \log(T)})$的界。本文聚焦于带度量动作空间的老虎机问题,通过链式论证,为一种Thompson Sampling变体建立了依赖于动作空间度量熵的新界。在奖励满足适当连续性假设的条件下,本文的界为d维线性老虎机问题给出了$O(d\sqrt{T})$的紧速率。

英文摘要:

This paper studies the Bayesian regret of a variant of the Thompson-Sampling algorithm for bandit problems. It builds upon the information-theoretic framework of [Russo and Van Roy, 2015] and, more specifically, on the rate-distortion analysis from [Dong and Van Roy, 2020], where they proved a bound with regret rate of $O(d\sqrt{T \log(T)})$ for the $d$-dimensional linear bandit setting. We focus on bandit problems with a metric action space and, using a chaining argument, we establish new bounds that depend on the metric entropy of the action space for a variant of Thompson-Sampling. Under suitable continuity assumption of the rewards, our bound offers a tight rate of $O(d\sqrt{T})$ for $d$-dimensional linear bandit problems.

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。

↑