Closed-Loop Multi-Property Materials Discovery, Progress and Opportunities

Machine learning (ML) techniques present tremendous opportunities to accelerate materials design and discovery, but significant developments are required to build on approaches from other domains. For example, ML models for materials must generally contend with sparser and more inhomogeneous data, these challenges compound in practical tasks that require simultaneous optimization of multiple properties. We present results from a “closed-loop” approach that integrates ML model predictions with experimental synthesis and characterization to provide new data and update the models. We first demonstrated success in the discovery of superconducting compounds by utilizing information from several public databases with in-house synthesis and characterization of crystal structure and critical temperature to train our ML models. Building on that framework, we expanded the task to include prediction of mechanical properties, as characterized experimentally using high-throughput nanoindentation. Finally, we explore the promise of ML techniques applied toward practical materials discovery; targeting multiple properties simultaneously, and outlining some of the near term opportunities.