ERL FISH – Lingchen Zhu & Peter Tilke (Schlumberger-Doll Research)

Title: Accelerating Geoscientific AI with Synthetic Data

Abstract: Data is crucial for implementing, training and applying AIML solutions for subsurface interpretation. Synthetic data can generate labeled realistic data efficiently and at scale for AIML applications. It is non-identifiable so privacy and data security concerns don’t apply. Can be used to train and test AIML models, for exploratory data analysis, to validate assumptions, demonstrate results that can be obtained with AIML models, and reject models producing poor results without the cost of acquiring and incorporating real data. Can be used for transfer learning. In this session we would like to share our experiences with generating synthetic geoscientific data (specifically stratigraphic forward models) for the purpose of accelerating AIML workflows. Building a synthetic data library of geologic analogs represents a level of complexity which involves creating, managing and accessing vast amounts of data efficiently for a variety of use cases.

About this series:  MIT Earth Resources Laboratory’s Friday  Informal Seminar features guest speakers from industry and academia on  topics relevant to our lab, including geophysics, seismology, rock  physics, imaging, inversion, machine learning, and the energy industry.  Titles and abstracts will be posted here when available. Contact  fish_seminar_organizers@mit.edu for more information and Zoom password.