Fighting Contextual Bandits with Stochastic Smoothing

Abstract

We introduce a new stochastic smoothing perspective to study adversarial contextual bandit problems. We propose a general algorithm template that represents random perturbation based algorithms and identify several perturbation distributions that lead to strong regret bounds. Using the idea of smoothness, we provide an $O(\sqrt{T})$ zero-order bound for the vanilla algorithm and an $O(L_T^{\ast 2/3})$ first-order bound for the clipped version. These bounds hold when the algorithms use with a variety of distributions that have a bounded hazard rate. Our algorithm template includes EXP4 as a special case corresponding to the Gumbel perturbation. Our regret bounds match existing results for EXP4 without relying on the specific properties of the algorithm.

merged to a manuscript "Online Learning via the Differential Privacy Lens," which can be found here: arXiv:1711.10019