Adaptive mobile behavior change intervention using reinforcement learning

Lihua Cai, Congyu Wu, Kiana Jafari Meimandi, Matthew S. Gerber

Abstract: As smartphones become increasingly intimate and continuous companions, many opportunities are arising in human behavior sensing, modeling, and coaching. This position paper explores opportunities and challenges for mobile-based deployment of behavior change interventions. We suggest the adoption and extension of reinforcement learning for addressing these challenges, and we identify several key areas of future research that, on the basis of prior results, appear ripe for extending the benefits of reinforcement learning to human behavior change. These areas include stronger grounding of states in theories of human behavior, RL agent adaptation and decomposition, cooperative reinforcement learning, and in situ evaluation.