Reproduce “Geophysical inverse problems with measurement-guided diffusion models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Matteo Ravasi
This study analyzes and compares measurement-guided diffusion models, Diffusion Posterior Sampling (DPS) and Pseudo-inverse Guided Diffusion Model (PGDM), for solving geophysical inverse problems. It demonstrates that careful problem re-parametrization and the selection of training data are critical for achieving improved solution quality and robust uncertainty quantification on synthetic and field datasets.
Key insights
- Effective re-parametrization of geophysical problems is crucial for successful application of diffusion models, adapting data to the required -1 to 1 amplitude range.
- The choice and characteristics of the training dataset used to learn the implicit prior have a more substantial impact on the quality and geological realism of inverse solutions and uncertainty estimates than the specific sampling algorithm itself.
- Omitting the denoiser's Jacobian in the PGDM update can surprisingly boost performance for seismic interpolation, and a 'gapped reverse diffusion process' can further enhance reconstruction quality by reducing sampling steps.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.