I will give an overview on how techniques from neural (conditional) density estimation and surrogate modeling can be used to solve challenging problems in seismic imaging and monitoring of geological carbon storage. I will start by outlining how (conditional) normalizing flows can be used as priors, to regularize inverse problems, and as low-fidelity amortized posteriors for wave-based inversions. To this end, I will use techniques from simulation-based inference. When time permits I will also talk about permeability inversion from time-lapse seismic data using neural surrogates (Fourier Neural Operators) to mimic solution operators of two-phase flow equations.

Pizza lunch will be served.

We gratefully acknowledge generous financial support by the Pacific Institute for the Mathematical Sciences (PIMS) and the Institute of Applied Mathematics (IAM).