Lecturer: Dr. Sudipto Banerjee, University of California Los Angeles

 

Title: Artificially Intelligent Geospatial Systems: A Case Study in Spatial Energetics for Mobile Health Data

Abstract: I will share my perspectives on the significant paradigm shift taking place in data analysis with the advent of AI technologies. This rapidly evolving field offers substantial intellectual space for statistical theory and methods to not only co-exist with other disciplines within computer science and machine learning, but also play a crucial role in advancing data analysis and probabilistic inference at unprecedented scales. I will elucidate three ideas that will synthesize into an artificially intelligent inferential system. The first is "amortized Bayesian inference" that considers training and calculating posterior distributions using generative AI. The second is Bayesian transfer learning for scaling Inference to massive datasets. The third is Bayesian predictive stacking that delivers exact simulation-based inference without resorting to expensive iterative methods such as Markov chain Monte Carlo. These ideas will be synthesized to offer a real time analysis of a case study in spatial energetics. Spatial energetics is broadly referred to as the study of live movement in real time and has been especially relevant in the context of mobile health data using actigraph units embedded in wearable devices. The case study is a part of the University of California Los Angeles Physical Activity and Sustainable Transportation Approaches (PASTA-LA) being conducted by the UCLA Fielding School of Public Health and is primarily concerned with learning about a subject's metabolic levels as a function of their mobility attributes and other health attributes.