Flow Matching Applied to Geophysical Probabilistic Inversion.

Probabilistic inversion is a critical technique in geophysics for understanding subsurface structures, but it often involves complex and computationally intensive methods. This research explores a novel application of Flow Matching, a technique originating from generative Artificial Intelligence, to address these challenges. The authors adapt the established mathematical framework of Flow Matching to the specific requirements of probabilistic inversion, particularly in the context of seismic Full-Waveform Inversion. The efficacy of this approach is demonstrated through two case studies: an initial simple 2D velocity model to illustrate its fundamental features, followed by an evaluation on the more complex OpenFWI dataset, showcasing its potential for real-world seismic velocity model inversion. This work highlights a new avenue for leveraging generative AI in scientific inverse problems.