RoME: A Robust Mixed-Effects Bandit Algorithm for Optimizing Mobile Health Interventions

TL;DR

RoME is a novel robust mixed-effects bandit algorithm designed to improve mobile health intervention optimization by addressing participant heterogeneity, nonstationarity, and nonlinear relationships, with proven theoretical guarantees and superior empirical performance.

Contribution

Introduces RoME, a new bandit algorithm that models user and time effects, incorporates network cohesion, and uses debiased machine learning for flexible reward estimation, advancing mobile health personalization.

Findings

RoME achieves robust regret bounds dependent only on model dimension.

Demonstrates superior performance in simulations and off-policy evaluations.

Effectively handles heterogeneity, nonstationarity, and nonlinear reward relationships.

Abstract

Mobile health leverages personalized and contextually tailored interventions optimized through bandit and reinforcement learning algorithms. In practice, however, challenges such as participant heterogeneity, nonstationarity, and nonlinear relationships hinder algorithm performance. We propose RoME, a Robust Mixed-Effects contextual bandit algorithm that simultaneously addresses these challenges via (1) modeling the differential reward with user- and time-specific random effects, (2) network cohesion penalties, and (3) debiased machine learning for flexible estimation of baseline rewards. We establish a high-probability regret bound that depends solely on the dimension of the differential-reward model, enabling us to achieve robust regret bounds even when the baseline reward is highly complex. We demonstrate the superior performance of the RoME algorithm in a simulation and two off-policy evaluation studies.