On the Approximation Accuracy of Gaussian Variational Inference

TL;DR

This paper analyzes the approximation accuracy of Gaussian Variational Inference, providing bounds on total variation and mean-covariance errors in high-dimensional Bayesian inference.

Contribution

It offers the first theoretical bounds on the approximation errors of Gaussian VI, relating them to dimension and sample size.

Findings

Bounds on TV error and mean-covariance error derived

Error analysis based on Hermite series expansion of the log posterior

Provides insights into the limitations and accuracy of Gaussian VI

Abstract

The main computational challenge in Bayesian inference is to compute integrals against a high-dimensional posterior distribution. In the past decades, variational inference (VI) has emerged as a tractable approximation to these integrals, and a viable alternative to the more established paradigm of Markov Chain Monte Carlo. However, little is known about the approximation accuracy of VI. In this work, we bound the TV error and the mean and covariance approximation error of Gaussian VI in terms of dimension and sample size. Our error analysis relies on a Hermite series expansion of the log posterior whose first terms are precisely cancelled out by the first order optimality conditions associated to the Gaussian VI optimization problem.