We are pleased to invite you to our February event with Dr. Gabrio Rizzuti from Shearwater. The event will take place in the faculty of Civil Engineering and Geoscience at TU Delft. The exact room will be shared in this document in the future.

This event is sponsored by the Delphi Consortium

Event's agenda:

17:00 - 17:10: Welcome and introduction by Dr. Joeri Brackenhoff

17:10 - 17:40: Technical talk by Dr. Gabrio Rizzuti on:

"Towards scalable uncertainty quantification for seismic inverse problems with deep learning"

17:40 - 18:00: Q&A and discussion

18:00 -            : Closing and drinks

Abstract:
In seismic inverse problems, the noise and illumination configuration severely impact the interpretation of the subsurface. It is natural to ask to what extent a given result can be perturbed while still fitting the observations, thus producing a "credibility" score. This is, in essence, the goal of uncertainty quantification (UQ)! Unfortunately, conventional sampling-based UQ methods, such as Markov chain Monte Carlo, are not adequate for industrial-scale problems, which consist of billions of unknowns. In this talk, we discuss a scalable solution based on special invertible neural networks and the way forward for 3D industrial applications. 

About Gabrio Rizzuti: 

Gabrio Rizzuti is a Senior Research Geophysicist at Shearwater GeoServices. His research interests mostly focused on seismic inverse problems, with emphasis on numerical modeling, FWI, and deep-learning-based uncertainty quantification. He's a background in pure and applied Maths, holds a PhD at Delft University of Technology, and he worked as a research scientist at Georgia Tech and Utrecht University.