InterPore2021

Generating unrepresented geological realizations using Generative Adversarial Networks

3 Jun 2021, 20:00

1h

Poster (+) Presentation (MS15) Machine Learning and Big Data in Porous Media Poster +

Speaker

Mr Alhasan Abdellatif (Heriot-Watt university)

Description

In this work, we investigate the capacity of Generative Adversarial
Networks (GANs) in generating unrepresented patterns in a geological
dataset. The new unrepresented patterns in the training dataset are as-
sumed to belong to the same original data distribution. Speci?cally, we
design a conditional GANs model in a supervised training of GANs, to in-
terpolate geological properties between the training classes. The presented
study includes an investigation of various training settings and model ar-
chitectures. In addition, we devised new conditioning routines, for an
improved generation of the missing samples. The presented numerical ex-
periments on images representing binary channels showed good geological
consistency as well as strong correlation with the target conditions.