(Nearly) Optimal Differentially Private Stochastic Multi-Arm Bandits

We study the problem of private stochastic multi-arm bandits. Our notion of privacy is the same as some of the earlier works in the general area of private online learning [13, 17, 24]. We design algorithms that are i) differentially private, and ii) have regret guarantees that (almost) match the regret guarantees for the best non-private algorithms (e.g., upper confidence bound sampling and Thompson sampling). Moreover, through our experiments, we empirically show the effectiveness of our algorithms.