Bayesian optimal experimental design in inverse problems
Conference
65th ISI World Statistics Congress
Format: IPS Abstract - WSC 2025
Keywords: design of experiment, statistical workflow, experimental planning
Abstract
Data collection, whether in field experiments or in the laboratory, is often restricted by limited resources. It can be difficult, expensive and time-consuming, which puts severe limits on the quality of data acquired. Bayesian inference provides a principled approach to quantify and optimise the information gained from any given experimental setup. However, the required computational effort in large-scale inverse problems still remains prohibitive. In this talk I discuss stability of Bayesian optimal experimental design (OED) under perturbation such as surrogate models or the prior. Additionally, I explore novel utility concepts that exhibit preferred stability properties.