Multilevel Gaussian Processes with Applications to Inverse Problems
Abstract
We present a multilevel Gaussian process (GP) regression framework for approximating functions that are accessible only through a hierarchy of approximations of increasing accuracy and computational cost. One example is the response surface of a parametrized partial differential equation, where the hierarchy arises from the resolution of the numerical solver. Modeling level increments independently with Matérn-kernel GPs and allocating samples flexibly across discretization levels yields substantial computational savings over single-level GP regression. We derive L2 convergence rates for the posterior mean and variance, accounting for maximum likelihood estimation of the marginal variance and for kernel misspecification. When increment regularity or solver cost is unknown a priori, an adaptive sequential-design algorithm with cost-adapted active learning criteria allocates samples across levels. We apply the surrogate to Bayesian inverse problems, for which Hellinger distance bounds between the true and approximate posteriors are known. Numerical experiments illustrate the theoretical rates and the resulting computational savings.
Related
- Sequential Entropic Optimal Transport for Bayesian Inverse Problems via Multilevel Monte Carlo
- Posterior Consistency under dependent errors in non-linear inverse problems with applications to time evolution equations
- New hierarchical matrices for fast kernel matrix-vector products with Applications to Gaussian processes and kernel ridge regression
- Simultaneous Pointwise Majorization for Mixed Tail Processes with Applications in Gaussian Chaos and Ergodic Diffusions
- Multilevel limits of spiked random matrix corners processes
BibTeX
@misc{salatovic2026multilevel,
title = {{Multilevel Gaussian Processes with Applications to Inverse Problems}},
author = {Stjepan Salatovic and Aretha L. Teckentrup and Sebastian Krumscheid},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.23202},
eprint = {2610.23202},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.23202},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Multilevel Gaussian Processes with Applications to Inverse Problems
AU - Stjepan Salatovic
AU - Aretha L. Teckentrup
AU - Sebastian Krumscheid
PY - 2026
DA - 2026/10/16
PB - arXiv
DO - 10.48550/arxiv.2610.23202
UR - https://arxiv.org/abs/2610.23202
AB - We present a multilevel Gaussian process (GP) regression framework for approximating functions that are accessible only through a hierarchy of approximations of increasing accuracy and computational cost. One example is the response surface of a parametrized partial differential equation, where the hierarchy arises from the resolution of the numerical solver. Modeling level increments independently with Matérn-kernel GPs and allocating samples flexibly across discretization levels yields substantial computational savings over single-level GP regression. We derive L^2 convergence rates for the posterior mean and variance, accounting for maximum likelihood estimation of the marginal variance and for kernel misspecification. When increment regularity or solver cost is unknown a priori, an adaptive sequential-design algorithm with cost-adapted active learning criteria allocates samples across levels. We apply the surrogate to Bayesian inverse problems, for which Hellinger distance bounds between the true and approximate posteriors are known. Numerical experiments illustrate the theoretical rates and the resulting computational savings.
ER -