In this work, we propose a deep learning pipeline for training neural networks that can approximate attributes in three dimensional seismic datasets. Following a supervised learning setting, we use generative adversarial modeling and train specialized networks that learn to translate input amplitude volumes into seismic attributes. Trained networks compute transformations much faster than attributes exact formulation since inference time is highly optimized and rapid in modern GPU architectures. Initial results show that conditional GANs are robust to learn how to compute different attributes with distinct data distributions, without any architecture modification. Via model inference, attribute computations are up to 80x faster than classical formulation.
Presentation Date: Monday, October 12, 2020
Session Start Time: 1:50 PM
Presentation Time: 3:05 PM
Location: Poster Station 1
Presentation Type: Poster
