arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1902.10089cs.LGstat.ML
收藏

随机多臂强盗中的扰动历史探测

Perturbed-History Exploration in Stochastic Multi-Armed Bandits

Google Research(谷歌研究院);DeepMind(深度思考);University of Alberta(阿尔伯塔大学);Facebook AI Research(脸书人工智能研究院)

Branislav Kveton, Csaba Szepesvari, Mohammad Ghavamzadeh, Craig Boutilier

摘要:

提出了一个随机多臂强盗中最小化累积后悔的在线算法。该算法将$O(t)$ i.i.d.在回合$t$中伪奖励其历史,然后拉动具有其扰动历史中最高平均奖励的手臂。因此,我们称之为扰动历史探测(PHE)。伪奖励是精心设计的,以抵消潜在低估的平均奖励的武器具有很高的概率。我们推导出了PHE的$n$ -轮遗憾的接近最优的间隙相关和无间隙的界限。我们分析的关键步骤是一个新的论点,表明随机伯努利奖励导致乐观。最后,我们实证评估PHE,并表明它是具有竞争力的国家的最先进的基线。

英文摘要:

We propose an online algorithm for cumulative regret minimization in a stochastic multi-armed bandit. The algorithm adds $O(t)$ i.i.d. pseudo-rewards to its history in round $t$ and then pulls the arm with the highest average reward in its perturbed history. Therefore, we call it perturbed-history exploration (PHE). The pseudo-rewards are carefully designed to offset potentially underestimated mean rewards of arms with a high probability. We derive near-optimal gap-dependent and gap-free bounds on the $n$-round regret of PHE. The key step in our analysis is a novel argument that shows that randomized Bernoulli rewards lead to optimism. Finally, we empirically evaluate PHE and show that it is competitive with state-of-the-art baselines.

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。

↑