New Optimization Paradigms for Large-Scale Probabilistic Inference
Probabilistic methods are ubiquitous across artificial intelligence, science, and engineering. On the one hand, probabilistic methods capture the central role of uncertainty, leading to high quality estimates and predictions in practice. On the other hand, modern applications to computer vision, bioscience, and robotics, among others, routinely require large complex models with millions of parameters which must be estimated using terabytes of data. Unfortunately, classic probabilistic estimation techniques simply do not accurately scale to such modern big data applications. During her CAS appointment, Professor He will address these foundational issues based on new optimization insights, with a view towards enabling a variety of high-impact applications that are not currently feasible. This project bridges large-scale optimization and probabilistic modeling and estimation with immediate high-impact applications to healthcare analytics, computer vision, and a wide spectrum of future applications to all areas of A.I.