Accelerating Tomographic Updates with Physics-Based Deep Learning and Realistic Earth Models

Paper submitted to IMAGE 2026, by Mark Roberts, Leandro Gabioli, Olga Brusova, David Brookes, Alejandro Valenciano (TGS).

Abstract

Velocity Model Building (VMB) remains a significant bottleneck in frontier exploration due to the intensive iterative nature of tomography and the frequent absence of high-quality initial models. We present a pragmatic approach to automating tomographic updates (MLTomo) using advanced neural architectures and physics-based regularization. This study seeks to close the gap between aspirational machine learning research and production-ready workflows that generalize on real-world datasets.

Read the full article here.