Abstract:SIAM Journal on Imaging Sciences, Volume 17, Issue 1, Page 301-333, March 2024. In this work, a method for obtaining pixelwise error bounds in Bayesian regularization of inverse imaging problems is introduced. The proposed method employs estimates of the posterior variance together with techniques from conformal prediction in order to obtain coverage guarantees for the error bounds, without making any assumption on the underlying data distribution. It is generally applicable to Bayesian regularization approaches, independent, e.g., of the concrete choice of the prior. Furthermore, the coverage guarantees can also be obtained in case only approximate sampling from the posterior is possible. With this in particular, the proposed framework is able to incorporate any learned prior in a black-box manner. Guaranteed coverage without assumptions on the underlying distributions is only achievable since the magnitude of the error bounds is, in general, unknown in advance. Nevertheless, experiments with multiple regularization approaches presented in the paper confirm that, in practice, the obtained error bounds are rather tight. For realizing the numerical experiments, a novel primal-dual Langevin algorithm for sampling from nonsmooth distributions is also introduced in this work, showing promising results in practice. While a proof of convergence for this primal-dual algorithm is still open, the theoretical guarantees of the proposed method do not require a guaranteed convergence of the sampling algorithm.

Statistical guarantees for Bayesian uncertainty quantification in non-linear inverse problems with Gaussian process priors

On some information-theoretic aspects of non-linear statistical inverse problems

On the Frequentist Coverage of Bayes Posteriors in Nonlinear Inverse Problems

A Bayesian approach with Gaussian priors to the inverse problem of source identification in elliptic PDEs

Bernstein - von Mises theorems for statistical inverse problems I: Schrödinger equation

Variational Gaussian Processes For Linear Inverse Problems

Nonlinear Expectation Inference for Direct Uncertainty Quantification of Nonlinear Inverse Problems

A posterior contraction for Bayesian inverse problems in Banach spaces

Consistency of variational Bayesian inference for non-linear inverse problems of partial differential equations

Consistency of Bayesian inference with Gaussian process priors in an elliptic inverse problem

Uncertainty Quantification for Sparse Deep Learning

Semi-parametric Bernstein-von Mises in Linear Inverse Problems

Generalized Bayes Approach to Inverse Problems with Model Misspecification

Bernstein -- von Mises theorems for statistical inverse problems II: Compound Poisson processes

Inverse Problems in a Bayesian Setting

Consistency of Bayesian inference with Gaussian process priors for a parabolic inverse problem

On the Bernstein-Von Mises Theorem for High Dimensional Nonlinear Bayesian Inverse Problems

Numerical posterior distribution error control and expected Bayes Factors in the bayesian Uncertainty Quantification of Inverse Problems

Posterior-Variance–Based Error Quantification for Inverse Problems in Imaging

Introduction To Gaussian Process Regression In Bayesian Inverse Problems, With New ResultsOn Experimental Design For Weighted Error Measures

Semiparametric inference using fractional posteriors