Orthogonally Decoupled Variational Fourier Features

TL;DR

This paper introduces an orthogonally decoupled variational basis that combines spectral methods with sparse inducing points to improve Gaussian process approximation, demonstrating competitive performance on various datasets.

Contribution

It presents a novel method that integrates spectral and sparse inducing point techniques using orthogonal decoupling, enhancing Gaussian process modeling.

Findings

Competitive performance on synthetic data

Effective on real-world datasets

Combines spectral and sparse methods efficiently

Abstract

Sparse inducing points have long been a standard method to fit Gaussian processes to big data. In the last few years, spectral methods that exploit approximations of the covariance kernel have shown to be competitive. In this work we exploit a recently introduced orthogonally decoupled variational basis to combine spectral methods and sparse inducing points methods. We show that the method is competitive with the state-of-the-art on synthetic and on real-world data.