Machine learning-based discovery of molecules, crystals, and composites: A perspective review

Abstract

Machine learning based approaches to material discovery are reviewed with the aim of providing a perspective on the current state of the art and its potential. Various models used to represent molecules and crystals are introduced and such representations can be used within the neural networks to generate materials that satisfy specified physical features and properties. For problems where large database for structure-property map cannot be created, the active learning approaches based on Bayesian optimization to maximize the efficiency of a search are reviewed. Successful applications of these machine learning based material discovery approaches are beginning to appear and some of the notable ones are reviewed.