Segmenting and Labeling On-Body Sensor Data Streams with CRFs and Factor Graphs

Ubiquitous mobile physiological sensing has the potential to profoundly improve our understanding of human health and behavior and to inform the design of more targeted treatments for a wide variety of conditions. However, extracting useful knowledge about high-level behaviors and activities from noisy low-level sensor data is an extremely challenging problem. In this talk, I will present two projects focused on segmenting and labeling on-body sensor data streams. The first project leverages dynamically generated conditional random field (CRF) models to parse wireless ECG data as a basic building block in a system designed to detect cocaine usage through its effect on ECG morphological structure. The second project is based on a novel and more general factor graph model for hierarchical segmentation and labeling that is being applied to a variety of domains including the detection and delineation of episodes of smoking and eating based on respiration and gestural data.
I am an assistant professor in the School of Computer Science at the University of Massachusetts Amherst where I co-direct the Machine Learning for Data Science Lab with Brendan O'connor and Dan Sheldon. I was previously a fellow of both the Pacific Institute for the Mathematical Sciences and the Killam Trusts at the University of British Columbia where I was based in the Laboratory for Computational Intelligence in the Department of Computer Science. I completed my PhD in machine learning in the Department of Computer Science at the University of Toronto.