Modeling Social Network Dynamics using Sensor Data
The structure and dynamics of social networks are of critical importance to many social phenomena, ranging from organizational efficiency to the spread of knowledge and disease. However, substantial advances in modeling the dynamics of social networks have been frustrated by the paucity of data rich enough for empirical investigation. A deeper understanding of people’s interactions will have significant impacts, not only on the social science research literature, but also by adding value to applied endeavors such as designing public spaces and office environments, and developing computer collaboration tools. We are investigating an innovative approach for learning the structure and dynamics of social networks, namely using sensing and communications tools pulled from the ubiquitous computing, coupled with machine learning techniques to unobtrusively study large populations of interacting humans over extended periods of time.
This is a joint project with University of Washington and Intel Seattle and has been funded by NSF.
I co-organized (with James Kitts) a multi-disciplinary workshop on Modeling Social Dynamics that was sponsored by NSF.
PIs: Tanzeem Choudhury, Jeff Bilmes, James Kitts, Henry Kautz, and Dieter Fox
Graduate Student: Danny Wyatt and Andrew Guillory
Learning Human Behavior from Multi-modal Data
Learning People's Routine and Rituals
In recent years people have started to explore sensor modalities (e.g. accelerometer, RFID) in addition to audio and video for recognizing and modeling human
behavior and activity. We are developing a probabilistic framework that allows us to systematically select and combine the most useful set of features from different modalities and recognize of a wide range of human activities.
Collaborators: Lin Liao, Jonathan Lester, Gaetano Borriello, Matthai Philipose, Dieter Fox, and Jeff Bilmes
Improving Sensor-based Activity Models using Automatically Mined Common-sense Priors
In this project we are exploring ways of combining commonsense knowledge and information mined from the web with sensor data collected from humans. Can we use relational information about the world to learn models of activity more efficiently and develop models that generalize well to novel data?
Collaborator: Matthai Philipose, Emmanuel Munguia-Tapia and Danny Wyatt
Please see my publications for more information and to download papers.