Distributionally Robust Gaussian Process Regression and Bayesian Inverse Problems

TL;DR

This paper introduces a distributionally robust Bayesian approach for Gaussian process regression and inverse problems, ensuring optimal predictions against worst-case models within a Wasserstein ball, with theoretical guarantees and practical algorithms.

Contribution

It formulates a novel min-max optimization framework for robust Bayesian nonparametric estimation, proving strong duality and existence of a Gaussian worst-case distribution.

Findings

The game has a well-defined value with strong duality.

A unique Nash equilibrium exists and can be approximated finitely.

Numerical experiments demonstrate the method's robustness and versatility.

Abstract

We study a distributionally robust optimization formulation (i.e., a min-max game) for two representative problems in Bayesian nonparametric estimation: Gaussian process regression and, more generally, linear inverse problems. Our formulation seeks the best mean-squared error predictor, in an infinite-dimensional space, against an adversary who chooses the worst-case model in a Wasserstein ball around a nominal infinite-dimensional Bayesian model. The transport cost is chosen to control features such as the degree of roughness of the sample paths that the adversary is allowed to inject. We show that the game has a well-defined value (i.e., strong duality holds in the sense that max-min equals min-max) and that there exists a unique Nash equilibrium which can be computed by a sequence of finite-dimensional approximations. Crucially, the worst-case distribution is itself Gaussian. We explore properties of the Nash equilibrium and the effects of hyperparameters through a set of numerical experiments, demonstrating the versatility of our modeling framework.