Hosted by Jemima Tabeart

Abstract

Inverse problems are prevalent across various fields of science and engineering, with applications spanning medical imaging, materials science, nondestructive testing, astrophysics, climate science, and seismology. These problems share a common goal: to estimating a quantity of interest from measurements obtained under specific experimental conditions. Such problems are typically solved with an iterative procedure that fits simulated to measured data. Uncertainty estimates of the estimate can be obtained by sampling a pre-defined posterior distribution. Recent advances in (conditional) generative models offer a promising approach to combine simulation, inference, and uncertainty quantification in a single data-driven model. This talk reviews some of these advancements in the context of linear inverse problems, focusing on the use of simple generative models for inference and uncertainty quantification in computed tomography.