Statistics and Data Science Seminar: Gaussian Process Computing with Vecchia's Approximation and the GpGp R Package

Co-sponsored by TRIADS

Abstract: Introduced more than 30 years ago, Vecchia's Gaussian process approximation has recently gained widespread popularity and is now emerging as a state-of-the-art approximation. I will review some of our and others' work on the approximation. This includes methods for improving the approximation by reordering the input locations, a grouped version for increasing accuracy and computational speed, theoretical developments putting other popular approximations in context relative to Vecchia's approximation, and procedures for fast optimization of the likelihood. We will also showcase the implementation of these ideas in the GpGp R package and highlight some applied projects that make use of the approximation for analyzing remote sensing data.

Bio: Joe Guinness is an Associate Professor in the Department of Statistics and Data Science at Cornell University. His research is focused on Gaussian process models, particularly theory for model development, spectral and sparse methods for efficient computation, and applications in Earth sciences and remote sensing. He teaches courses in statistical computation and linear models at various levels. Guinness received his undergraduate degree in mathematics and physics from Washington University in St. Louis in 2007 and his Ph.D. in statistics from the University of Chicago in 2012.

Host: Xuming He