Dissertation Proposal Defense - Orson Xu
Interpretable, Personalized, Generalizable Behavior Modeling for Longitudinal Health and Well-being
Abstract: As artificial-intelligent-powered devices become more embedded in our lives, they offer an unprecedented ability to passively sense our daily behavior at a high resolution. An essential capability for these devices is to understand and model the behavior of interest for various aspects of human health and well-being. There is a growing body of research on behavior modeling via passive sensing data to predict specific behavioral outcomes, such as detecting physical health issues and monitoring mental health status. However, due to the scarcity of the behavior label, existing machine learning techniques for behavior modeling still have a set of challenges for robust real-life deployment: interpretability, personalization, and generalizability. Meaningful and effective support of end-user goals requires interpretability to achieve a transparent, trustful, and deployable system. Moreover, each individual has their own unique behavior pattern, ability, and preference, making personalization important for taking individual differences into account, so that predictions and interventions can be tailored to each individual. Addressing these two challenges in a single dataset is the prerequisite for building deployable and generalizable modeling techniques. In real-life deployment, the behavior patterns of new users and contexts could have a different distribution. This requires a technique with cross-dataset generalizability. My research aims to solve the three challenges by collecting passive sensing datasets and developing new machine learning algorithms. I have made progress in tackling each of these challenges separately. I propose to develop a better algorithm to address these three challenges simultaneously.
Chair: Anind Dey, Professor and Dean, Information School, UW
GSR: James Fogarty, Professor, Allen School of Computer Science and Engineering, UW
Member: Jennifer Mankoff, Richard E. Ladner Endowed Professor and Associate Director, Allen School of Computer Science and Engineering, UW
Member: Tim Althoff, Assistant Professor, Allen School of Computer Science and Engineering, UW
Member: Andrew T. Campbell, Albert Bradley 1915 Third Century Professor, Department of Computer Science, Dartmouth College