PHDL Seminar Series

Interpretable Spatiotemporal Learning to Forecast Human-Societal Activity

Monday 4/8 12:00PM - 12:00PM
1155 Public Health, Foster Conf. Rm.(former 1149)

The relationships between individual behavior and broader-scale societal structures have been central to a range of human phenomena, from job and migration decisions to political participation to the experience of a specific health-related event. Recent progress in predictive analytics along with big data offers a powerful way to explore those relationships but often gives rise to the "black box" problem with little insight into what goes on in the algorithmically learned relationships.  In this talk, Dr. Lin will present a spatiotemporal learning approach that leverages a deep learning framework with relevant social theories to help examine the relationship between human-societal activity and their social and geographical contexts, with applications including predicting political protest and opioid overdose events. The approach is not only capable of forecasting the occurrence of future events, but also provides theory-relevant interpretations -- it allows for interpreting what features, from which places, have significant contributions on the forecasting model, as well as how they make those contributions.

Last Updated On Tuesday, April 2, 2019 by Crow, Sharon Weber
Created On Tuesday, April 2, 2019