Conference Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 229 - Geostatistical Computing on Modern Parallel Architectures
Type: Topic Contributed
Date/Time: Tuesday, August 9, 2022 : 8:30 AM to 10:20 AM
Sponsor: Section on Statistical Computing
Abstract #320927
Title: Scalable Gaussian-Process Regression and Variable Selection Using Vecchia Approximations
Author(s): Jian Cao* and Matthias Katzfuss and Marc Genton and Joe Guinness
Companies: Texas A&M University and Texas A&M University and KAUST and Cornell University
Keywords: adaptive bridge penalty; constrained coordinate descent; gradient-based variable selection; mini-batch subsampling; ordered conditional approximation; penalized Gaussian regression
Abstract:

Gaussian process (GP) regression is a flexible, nonparametric approach to regression that naturally quantifies uncertainty. In many applications, the number of observations and covariates are both large, and a goal is to select covariates that are related to the response. For this setting, we propose a novel, scalable method, which optimizes a penalized GP log-likelihood based on the Vecchia GP approximation from spatial statistics. We traverse the regularization path from strong to weak penalization, sequentially adding candidate covariates based on the gradient of the log-likelihood and deselecting irrelevant covariates via a new quadratic constrained coordinate descent algorithm. We propose Vecchia-based mini-batch subsampling, which provides unbiased gradient estimators. The resulting procedure is scalable to millions of observations and tens of thousands of covariates. Theoretical analysis and numerical studies demonstrate the improved scalability and accuracy relative to existing methods.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2022 program