Riemannian Diagnostics for Decoder-Consistent Latent-Space Analysis in Subsurface Modeling

Anton Shirokov, Gleb Shishaev, Boris Merzlikin

Submission Type: I want my submission to be considered for poster only

Keywords: Earth science, variational autoencoders, latent space, Riemannian geometry, pullback metric, uncertainty quantification

TL;DR: We propose a geometry-aware QA framework for subsurface VAE models that uses pullback-metric diagnostics and uncertainty-aware reliability maps to flag unsafe latent-space interpolations and support safer scenario exploration.

Confirmation Of Submission Requirements: I submit an abstract. It uses the template provided on the submission page and is no longer than 2 pages.

Submission Number: 222