Variance-based sensitivity of Bayesian inverse problems to the prior distribution
arXiv:2310.18488 · doi:10.1615/Int.J.UncertaintyQuantification.2024051475
Abstract
The formulation of Bayesian inverse problems involves choosing prior distributions; choices that seem equally reasonable may lead to significantly different conclusions. We develop a computational approach to better understand the impact of the hyperparameters defining the prior on the posterior statistics of the quantities of interest. Our approach relies on global sensitivity analysis (GSA) of Bayesian inverse problems with respect to the hyperparameters defining the prior. This, however, is a challenging problem--a naive double loop sampling approach would require running a prohibitive number of Markov chain Monte Carlo (MCMC) sampling procedures. The present work takes a foundational step in making such a sensitivity analysis practical through (i) a judicious combination of efficient surrogate models and (ii) a tailored importance sampling method. In particular, we can perform accurate GSA of posterior prediction statistics with respect to prior hyperparameters without having to repeat MCMC runs. We demonstrate the effectiveness of the approach on a simple Bayesian linear inverse problem and a nonlinear inverse problem governed by an epidemiological model.
References in corpus (8)
- Scalable and efficient algorithms for the propagation of uncertainty from data through inference to prediction for large-scale problems, with application to flow of the Antarctic ice sheet
- On a generalization of the preconditioned Crank-Nicolson Metropolis algorithm
- Goal-Oriented Optimal Design of Experiments for Large-Scale Bayesian Linear Inverse Problems
- On the Brittleness of Bayesian Inference
- Generalization Bounds for Sparse Random Feature Expansions
- Hyper-Differential Sensitivity Analysis for Inverse Problems Constrained by Partial Differential Equations
- Extreme learning machines for variance-based global sensitivity analysis
- Sensitivity Analysis of the Information Gain in Infinite-Dimensional Bayesian Linear Inverse Problems