Improving interpretation and modelling techniques in geoscience using deep generative networks (GANs)

Abstract

Generative Adversarial Networks are a novel, data-efficient method by which models are trained to learn from a limited amount of labelled data and to generate synthetic data, which reduces the burden of manual annotation. This thesis examines the generative adversarial power of GANs in geoscience, exploring how they can be used to unravel the underlying geological patterns and geostatistical features inherent in the geoscience data.