MLLGOct 11, 2018

Fighting Contextual Bandits with Stochastic Smoothing

arXiv:1810.05188v2
Originality Incremental advance
AI Analysis

This work provides a theoretical framework for improving regret bounds in adversarial contextual bandits, but it is incremental as it matches existing results like EXP4.

The authors tackled adversarial contextual bandit problems by introducing a stochastic smoothing perspective and a general algorithm template, achieving an O(√T) zero-order bound and an O(L_T^{*2/3}) first-order bound for regret.

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 $O(L^{*2/3}_{T})$ 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.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes