Juan Romero, Wolfgang Heidrich, and Matteo Ravasi

 GitHub: https://github.com/DeepWave-KAUST/B-IntraSeismic-pub

B-IntraSeismic is an extension of the IntraSeismic framework to Bayesian seismic inversion. The core innovation lies in leveraging implicit neural representations and variational inference to predict the posterior distribution of subsurface properties. We introduce two approaches: B-IntraSeismic (BIS) and B-IntraSeismic with Conditional Normalizing Flows (BIS-Flow). Both methods are evaluated on synthetic and field datasets, demonstrating their ability to produce high-resolution mean models and standard deviation maps (mean-field approximation) with significantly lower computational costs compared to benchmark methods. This work highlights the potential of B-IntraSeismic to deliver scalable and accurate uncertainty quantification for seismic inversion.