|
Activity Number:
|
26
- Imaging Speed Session
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 8, 2021 : 1:30 PM to 3:20 PM
|
|
Sponsor:
|
Section on Statistics in Imaging
|
|
Abstract #318048
|
|
|
Title:
|
WITHDRAWN: Statistical Characterization of the Generative Adversarial Network (GAN) Modeling of Seismic Data
|
|
Author(s):
|
Bradley C Wallet and Joseph McNease
|
|
Companies:
|
Aramco Americas and University of Houston
|
|
Keywords:
|
Generative adversarial models;
Seismology;
Simulation;
Volumetric data;
Seismic imaging;
Geophysics
|
|
Abstract:
|
Generative Adversarial Networks (GANs) are a method for sampling from a high dimensional dataset with complex dependencies such as image data. In this method, two neural networks train in tandem. A generator network learns to sample from the unknown distribution while a discriminator learns to distinguish between real data samples and samples drawn from the generator network. Upon completion of training, the results are generally evaluated by examining resulting samples from the GAN and having an interpreter visually inspect them. Simple visual inspection may not validate the suitability of the generated data for use in machine learning workflows. Seismic data are complex, 3D volumetric data with significant economic value in the energy industry. This work demonstrates the capability to use GAN to sample from a distribution of seismic data. Seismic attributes are then generated using the generated images and real seismic data. The distributions of these attributes for the real and generated data are then compared to assess goodness of fit related to geological and geophysical processes. Implications for automated seismic interpretation workflows are then discussed.
|
Authors who are presenting talks have a * after their name.
