arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2103.02741cs.LG
收藏

没有全序武器的组合强盗

Combinatorial Bandits without Total Order for Arms

The University of Texas at Austin(得克萨斯大学奥斯汀分校);Amazon(亚马逊)

Shuo Yang, Tongzheng Ren, Inderjit S. Dhillon, Sujay Sanghavi

摘要:

我们考虑组合强盗问题,其中在每个时间步,在线学习者从臂集合$\mathcal{A}$中选择大小为$k$的子集$s$,其中$\left|\mathcal{A}\right| = n$,并观察所选集合$s$中每个臂的随机奖励。在线学习者的目标是最小化由于不选择最大化预期总回报的$s^*$而引起的遗憾。具体来说,我们关注一个具有挑战性的设置,其中1)手臂的奖励分配取决于它所属的集合$s$,至关重要的是2)$\mathcal{A}$中的手臂有\textit{no total order}。在本文中,我们正式提出了一个奖励模型,捕捉集依赖的奖励分布,并假设没有总的武器秩序。相应地,我们提出了一个置信上限(UCB)算法,保持UCB为每个单独的手臂,并选择与顶部$k$ UCB的武器。我们开发了一个新的遗憾分析,并显示了$O\left(\frac{k^2 n \log T}ε\right)$缺口依赖的遗憾界以及$O\left(k^2\sqrt{n T \log T}\right)$缺口独立的遗憾界。我们还提供了一个下界建议的奖励模型,这表明我们提出的算法是接近最优的任何常数$k$。在各种奖励模型上的实验结果证明了该算法的广泛适用性。

英文摘要:

We consider the combinatorial bandits problem, where at each time step, the online learner selects a size-$k$ subset $s$ from the arms set $\mathcal{A}$, where $\left|\mathcal{A}\right| = n$, and observes a stochastic reward of each arm in the selected set $s$. The goal of the online learner is to minimize the regret, induced by not selecting $s^*$ which maximizes the expected total reward. Specifically, we focus on a challenging setting where 1) the reward distribution of an arm depends on the set $s$ it is part of, and crucially 2) there is \textit{no total order} for the arms in $\mathcal{A}$. In this paper, we formally present a reward model that captures set-dependent reward distribution and assumes no total order for arms. Correspondingly, we propose an Upper Confidence Bound (UCB) algorithm that maintains UCB for each individual arm and selects the arms with top-$k$ UCB. We develop a novel regret analysis and show an $O\left(\frac{k^2 n \log T}ε\right)$ gap-dependent regret bound as well as an $O\left(k^2\sqrt{n T \log T}\right)$ gap-independent regret bound. We also provide a lower bound for the proposed reward model, which shows our proposed algorithm is near-optimal for any constant $k$. Empirical results on various reward models demonstrate the broad applicability of our algorithm.

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。

↑