Materials Science in the AI age: high-throughput library generation, machine learning and a pathway from correlations to the underpinning physics
Kamal Choudhary, Aaron G. Kusne, Francesca M. Tavazza, Jason R. Hattrick-Simpers, Rama K. Vasudevan, Apurva Mehta, Ryan Smith, Lukas Vlcek, Sergei V. Kalinin, Maxim Ziatdinov
The use of advanced data analytics, statistical and machine learning approaches (AI) to materials science has experienced a renaissance, driven by advances in computer sciences, availability and access of software and hardware, and a growing realization that data-driven methods can provide a new route to tackling age-old problems. In this prospective, we review some of the recent work on this topic, focusing on generation of libraries from both experiment and theoretical tools, across length scales. In each area, we highlight both the need for these libraries and the key advances facilitated by statistical and/or machine learning algorithms in providing new, previously unobtainable insights, and illustrate areas of improvement. We focus on the importance of community-driven efforts to build these libraries, and illustrate how modeling, macroscopic experiments and atomic-scale imaging can be combined to dramatically accelerate understanding and development of new material systems via a statistical physics framework. These point towards a data driven future wherein knowledge can be aggregated and used collectively, surpassing the capabilities of individual researchers, groups or institutions, and accelerating the advancement of materials science.