arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2205.09899stat.MLcs.AIcs.ITcs.LGmath.IT
收藏

打破$\sqrt{T}$障碍:随机上下文线性Bandits中的实例独立对数后悔

Breaking the $\sqrt{T}$ Barrier: Instance-Independent Logarithmic Regret in Stochastic Contextual Linear Bandits

AWS AI(亚马逊云科技人工智能部门)

Avishek Ghosh, Abishek Sankararaman

摘要:

我们证明了一个独立的实例(多)对数后悔随机上下文土匪线性回报。以前,在\cite{chu2011contextual},一个下界的$\mathcal{O}(\sqrt{T})$示出的上下文线性强盗问题与任意(adversarily选择)的背景。在本文中,我们表明,随机上下文确实有助于减少从$\sqrt{T}$到$\polylog(T)$的遗憾。我们提出了低遗憾随机上下文Bandits(\texttt{LR-SCB}),它利用随机上下文,并执行参数估计(在$\ell_2$范数)和遗憾最小化同时进行。\texttt{LR-SCB}以历元为单位工作,其中前一历元的参数估计用于减少当前历元的遗憾。\texttt{LR-SCB}的(多)对数后悔源于两个关键事实:(a)应用范数自适应算法来利用参数估计和(b)分析移位线性上下文强盗算法,表明移位导致后悔增加。我们还通过实验证明,随机背景确实会导致与$\polylog(T)$成比例的遗憾。

英文摘要:

We prove an instance independent (poly) logarithmic regret for stochastic contextual bandits with linear payoff. Previously, in \cite{chu2011contextual}, a lower bound of $\mathcal{O}(\sqrt{T})$ is shown for the contextual linear bandit problem with arbitrary (adversarily chosen) contexts. In this paper, we show that stochastic contexts indeed help to reduce the regret from $\sqrt{T}$ to $\polylog(T)$. We propose Low Regret Stochastic Contextual Bandits (\texttt{LR-SCB}), which takes advantage of the stochastic contexts and performs parameter estimation (in $\ell_2$ norm) and regret minimization simultaneously. \texttt{LR-SCB} works in epochs, where the parameter estimation of the previous epoch is used to reduce the regret of the current epoch. The (poly) logarithmic regret of \texttt{LR-SCB} stems from two crucial facts: (a) the application of a norm adaptive algorithm to exploit the parameter estimation and (b) an analysis of the shifted linear contextual bandit algorithm, showing that shifting results in increasing regret. We have also shown experimentally that stochastic contexts indeed incurs a regret that scales with $\polylog(T)$.

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。

↑