Deep Gaussian process priors for Bayesian image reconstruction
Open Access
- 11 June 2025
- journal article
- Published by IOP Publishing in Inverse Problems
- Vol. 41 (6) , 065016
Abstract
In image reconstruction, an accurate quantification of uncertainty is of great importance for informed decision making. Here, the Bayesian approach to inverse problems can be used: the image is represented through a random function that incorporates prior information which is then updated through Bayes’ formula. However, finding a prior is difficult, as images often exhibit non-stationary effects and multi-scale behavior. Thus, usual Gaussian process priors are not suitable. Deep Gaussian processes, on the other hand, encode non-stationary behavior in a natural way through their hierarchical structure. To apply Bayes’ formula, one commonly employs a Markov chain Monte Carlo (MCMC) method. In the case of deep Gaussian processes, sampling is especially challenging in high dimensions: the associated covariance matrices are large, dense, and changing from sample to sample. A popular strategy towards decreasing computational complexity is to view Gaussian processes as the solutions to a fractional stochastic partial differential equation (SPDE). In this work, we investigate efficient computational strategies to solve the fractional SPDEs occurring in deep Gaussian process sampling, as well as MCMC algorithms to sample from the posterior. Namely, we combine rational approximation and a determinant-free sampling approach to achieve sampling via the fractional SPDE. We test our techniques in standard Bayesian image reconstruction problems: upsampling, edge detection, and computed tomography. In these examples, we show that choosing a non-stationary prior such as the deep GP over a stationary GP can improve the reconstruction. Moreover, our approach enables us to compare results for a range of fractional and non-fractional regularity parameter values.Keywords
All Related Versions
Funding Information
- Engineering and Physical Sciences Research Council (EP/R014604/1)
This publication has 68 references indexed in Scilit:
- On Maximum a Posteriori Estimation with Plug & Play Priors and Stochastic Gradient DescentJournal of Mathematical Imaging and Vision, 2023
- On Randomized Trace Estimates for Indefinite Matrices with an Application to DeterminantsFoundations of Computational Mathematics, 2021
- Regularity and convergence analysis in Sobolev and Hölder spaces for generalized Whittle–Matérn fieldsNumerische Mathematik, 2020
- Krylov methods for inverse problems: Surveying classical, and introducing new, algorithmic approachesGAMM-Mitteilungen, 2020
- Fast random field generation with H-matricesNumerische Mathematik, 2018
- Optimal solvers for linear systems with fractional powers of sparse SPD matricesNumerical Linear Algebra with Applications, 2018
- A determinant‐free method to simulate the parameters of large Gaussian fieldsStat, 2017
- Randomized algorithms for generalized Hermitian eigenvalue problems with application to computing Karhunen–Loève expansionNumerical Linear Algebra with Applications, 2015
- Efficient approximation of random fields for numerical applicationsNumerical Linear Algebra with Applications, 2015
- Application of hierarchical matrices for computing the Karhunen–Loève expansionComputing, 2008