Scalable Variational Gaussian Process Classification

TL;DR

This paper presents a scalable variational Gaussian process classification method that efficiently handles millions of data points, outperforming existing techniques on benchmark datasets.

Contribution

It introduces a variational inducing point framework that significantly improves scalability and performance of Gaussian process classification.

Findings

Outperforms state-of-the-art on benchmark datasets

Enables classification on datasets with millions of points

Demonstrates practical scalability and effectiveness

Abstract

Gaussian process classification is a popular method with a number of appealing properties. We show how to scale the model within a variational inducing point framework, outperforming the state of the art on benchmark datasets. Importantly, the variational formulation can be exploited to allow classification in problems with millions of data points, as we demonstrate in experiments.