Contextual bandits with concave rewards, and an application to fair ranking

TL;DR

This paper introduces a novel algorithm for Contextual Bandits with Concave Rewards (CBCR) that achieves provably vanishing regret without policy restrictions, applicable to fair ranking and recommendation systems.

Contribution

It provides the first regret guarantees for CBCR without policy space restrictions and applies convex optimization techniques to multi-objective bandit problems.

Findings

First algorithm with vanishing regret for CBCR without policy restrictions

Applicable to linear and general reward functions in non-combinatorial actions

Includes a case study on fair ranking with regret guarantees

Abstract

We consider Contextual Bandits with Concave Rewards (CBCR), a multi-objective bandit problem where the desired trade-off between the rewards is defined by a known concave objective function, and the reward vector depends on an observed stochastic context. We present the first algorithm with provably vanishing regret for CBCR without restrictions on the policy space, whereas prior works were restricted to finite policy spaces or tabular representations. Our solution is based on a geometric interpretation of CBCR algorithms as optimization algorithms over the convex set of expected rewards spanned by all stochastic policies. Building on Frank-Wolfe analyses in constrained convex optimization, we derive a novel reduction from the CBCR regret to the regret of a scalar-reward bandit problem. We illustrate how to apply the reduction off-the-shelf to obtain algorithms for CBCR with both linear and general reward functions, in the case of non-combinatorial actions. Motivated by fairness in recommendation, we describe a special case of CBCR with rankings and fairness-aware objectives, leading to the first algorithm with regret guarantees for contextual combinatorial bandits with fairness of exposure.