Towards Future Health and Well-being: Bridging Behavior Modeling and Intervention

Abstract

With the advent of always-available, ubiquitous devices with powerful passive sensing and active interaction capabilities, the opportunities to integrate AI into this ecosystem have matured, providing an unprecedented opportunity to understand and support user well-being. A wide array of research has demonstrated the potential to detect risky behaviors and address health concerns, using human-centered ML to understand longitudinal, passive behavior logs. Unfortunately, it is difficult to translate these findings into deployable applications without better approaches to providing human-understandable relationship explanations between behavior features and predictions; and generalizing to new users and new time periods. My past work has made significant headway in addressing modeling accuracy, interpretability, and robustness. Moreover, my ultimate goal is to build deployable, intelligent interventions for health and well-being that make use of succeeding ML-based behavior models. I believe that just-in-time interventions are particularly well suited to ML support. I plan to test the value of ML for providing users with a better, interpretable, and robust experience in supporting their well-being.