Enhancing Full-Waveform Variational Inference through Stochastic Resampling

TitleEnhancing Full-Waveform Variational Inference through Stochastic Resampling
Publication TypePresentation
Year of Publication2024
AuthorsYunlin Zeng, Ziyi Yin, Rafael Orozco, Mathias Louboutin, Felix J. Herrmann
KeywordsAmortized Variational Inference, Bayesian inference, CIG, conditional normalizing flows, deep learning, FWI, Imaging, Inverse problems, MVA, RTM, Summary Statistics, Uncertainty quantification, WISE
Abstract

Recent developments in simulation-based inference, like the full-waveform variational inference via subsurface extensions (WISE), enable rapid online estimation of subsurface velocities by leveraging pre-trained models. To achieve this, WISE employs subsurface-offset common image gathers to convert shot data into physics-informed summary statistics. While common image gathers effectively retain critical information even when initial velocity estimates are inaccurate, WISE’s performance still depends on the assumption that the initial migration-velocity model is a single 1D velocity model. In this work, we present experiments using both 1D and 2D velocity models and develop a stochastic resampling method to generate variations of initial migration-velocity models. This technique allows us to systematically infer alternative velocity models that are consistent with the observed data, while enhancing the posterior sample quality and reducing dependency on the initial velocity model compared to the standard WISE approach.

URLhttps://slim.gatech.edu/Publications/Public/Conferences/ML4SEISMIC/2024/zeng2024ML4SEISMICefv
Citation Keyzeng2024ML4SEISMICefv