Fourier representations for fast Gaussian process regression

Data Science Seminar

Philip Greengard (Columbia University)

Abstract

Over the last couple of decades a large number of numerical methods have been introduced for efficiently performing Gaussian process regression. Most of these methods focus on fast inversion of the covariance matrix that appears in the Gaussian density. In this talk, I describe a slightly different approach to Gaussian process regression that relies on efficient weight-space representations of Gaussian processes. These representations have substantial advantages including computational benefits and model interpretability.