Recent progress in generative adversarial networks applied to inversely designing inorganic materials: A brief review

Abstract

Generative adversarial networks (GANs) are deep generative models (GMs) that have recently attracted attention owing to their impressive performance in generating completely novel images, text, music, and speech. Recently, GANs have made interesting progress in designing materials exhibiting desired functionalities, termed ‘inverse materials design’ (IMD). Because, discovering materials can lead to enormous technological progress, it is critical to provide a systematic review of new GAN applications to inversely designing inorganic materials. In this study, various aspects of GAN-based IMD were examined wherein IMD is a primary design process for discovering materials exhibiting desired features (physical properties, chemical formulae, etc.) by implementing constraints or conditions on input data or algorithms. We discussed fundamental materials databases and relevant machine-learning criteria. Furthermore, the comprehensive software tools currently available to materials scientists were presented. Descriptors including the criteria required for training GAN models were also discussed. Finally, we summarized both challenges and future direction for applying GANs to IMD research.