Sampling-Free Bayesian Inference for Local Refinement in Linear Inversion Problems With a Latent Target Property

We present a sampling free probabilistic inversion of latent target property based on the principles expectation propagation where we estimate joint distribution variable in local region. The prior model matches focused region but integrates out parameters outside focus using approximate distributions. includes large spatial structure information while maintaining dimension small. In addition, map and solve into new feature space can exclude components data has little influence, thereby decreasing dimensionality therefore runtime. test method seismic AVO examples for prediction facies classes, as well estimation vuggy porosity CT images core from carbonate reservoir. demonstrate that our achieves good quality predictions significantly reducing computational demand, making it particularly interesting to run scale studies.