A0308
Title: Scalable additive Gaussian process regression using Vecchia approximations
Authors: Isa Marques - The Ohio State University (United States) [presenting]
Paul Wiemann - The Ohio State University (United States)
Matthias Katzfuss - University of Wisconsin-Madison (United States)
Abstract: Generalized additive models (GAMs) have gained widespread popularity for their ability to link a set of covariates to a response through potentially non-linear effects. The recent integration of Gaussian processes (GPs) into this framework has further advanced modeling capability. In a configuration restricted to additive, one-dimensional GPs employing stationary and isotropic Matern covariance functions, it has been shown that the posterior mean and variance computations are achievable in quasilinear complexity. To overcome these restrictions while maintaining computational efficiency, a novel method is presented utilizing Vecchia approximations of the latent additive functions. This approach facilitates efficient computation of posterior mean and variance in more complex scenarios. Notably, it accommodates interactions and non-Matern covariance functions, which are pivotal in machine learning applications.
