Reinforcement Learning for Sustaining User Engagement in Digital Interventions

Monday, May 5, 2024 – 3:00 pm CT

About the Webinar:

We begin by introducing reinforcement learning (RL) and explaining its basic idea of learning by trial and error to optimize long-term outcomes. User engagement is critical to maintaining the effectiveness of digital interventions in mHealth. We then describe how RL can personalize the delivery of digital interventions to better match user needs and sustain engagement over time using conceptual models. After outlining the key concepts, we discuss some deployed algorithms in real-world mHealth studies and how engagement sustainability is achieved in practice. Specifically, we design RL agents either through reward construction that accounts for the delayed negative effects of digital interventions, as demonstrated in Oralytics, or by directly predicting the future delayed effects, as demonstrated in HeartSteps v2/v3. These design considerations discount the current benefits of digital interventions by acknowledging their potential negative delayed effects into the future, for example, through increased user app burdens, and thus prolong user engagement.

About the Presenter:

Ziping is a Postdoctoral Fellow in Statistics at Harvard University, working with Professor Susan Murphy. He earned his PhD in Statistics from the University of Michigan under the guidance of Professor Ambuj Tewari. His research focuses on sequential decision-making methods, such as Reinforcement Learning, with applications to digital intervention delivery in mobile health. He collaborates closely with domain experts to design and implement Reinforcement Learning agent in online mobile health clinical trials. More about Ziping Xu.