Optimal Learning for Structured Bandits

TL;DR

This paper introduces DUSA, a new algorithm for structured multi-armed bandits that exploits convex structural information to achieve near-optimal regret, outperforming classical methods that ignore such structure.

Contribution

The paper proposes DUSA, a novel, computationally feasible algorithm that matches the theoretical regret lower bound for various structured bandit problems, unifying and extending prior approaches.

Findings

DUSA achieves asymptotic minimal regret across multiple structured bandit settings.

It effectively exploits convex structural information to improve decision-making.

The approach unifies analysis for linear, Lipschitz, convex, and new structured bandits.

Abstract

We study structured multi-armed bandits, which is the problem of online decision-making under uncertainty in the presence of structural information. In this problem, the decision-maker needs to discover the best course of action despite observing only uncertain rewards over time. The decision-maker is aware of certain convex structural information regarding the reward distributions; that is, the decision-maker knows the reward distributions of the arms belong to a convex compact set. In the presence such structural information, they then would like to minimize their regret by exploiting this information, where the regret is its performance difference against a benchmark policy that knows the best action ahead of time. In the absence of structural information, the classical upper confidence bound (UCB) and Thomson sampling algorithms are well known to suffer minimal regret. As recently pointed out, neither algorithms are, however, capable of exploiting structural information that is commonly available in practice. We propose a novel learning algorithm that we call "DUSA" whose regret matches the information-theoretic regret lower bound up to a constant factor and can handle a wide range of structural information. Our algorithm DUSA solves a dual counterpart of the regret lower bound at the empirical reward distribution and follows its suggested play. We show that this idea leads to the first computationally viable learning policy with asymptotic minimal regret for various structural information, including well-known structured bandits such as linear, Lipschitz, and convex bandits, and novel structured bandits that have not been studied in the literature due to the lack of a unified and flexible framework.