6
26

Old Dog Learns New Tricks: Randomized UCB for Bandit Problems

Abstract

We propose RandUCB\tt RandUCB, a bandit strategy that builds on theoretically derived confidence intervals similar to upper confidence bound (UCB) algorithms, but akin to Thompson sampling (TS), it uses randomization to trade off exploration and exploitation. In the KK-armed bandit setting, we show that there are infinitely many variants of RandUCB\tt RandUCB, all of which achieve the minimax-optimal O~(KT)\widetilde{O}(\sqrt{K T}) regret after TT rounds. Moreover, for a specific multi-armed bandit setting, we show that both UCB and TS can be recovered as special cases of RandUCB\tt RandUCB. For structured bandits, where each arm is associated with a dd-dimensional feature vector and rewards are distributed according to a linear or generalized linear model, we prove that RandUCB\tt RandUCB achieves the minimax-optimal O~(dT)\widetilde{O}(d \sqrt{T}) regret even in the case of infinitely many arms. Through experiments in both the multi-armed and structured bandit settings, we demonstrate that RandUCB\tt RandUCB matches or outperforms TS and other randomized exploration strategies. Our theoretical and empirical results together imply that RandUCB\tt RandUCB achieves the best of both worlds.

View on arXiv
Comments on this paper