Abstract: The performance of many advanced technologies are inhibited by the slow process of optimizing materials. Physics-based theories and computational tools are proven routes to accelerating materials design but are unavailable or computationally intractable for many problems.
In this talk, we demonstrate how machine learning (ML) can close this capability gap. We start with a discussion of how to use ML on different types of materials data, with an emphasis on creating tools to replace costly density-functional theory calculations. We then will describe several case studies — including metallic glass design and radiation damage prediction — to show the advantages and pitfalls of using ML to design materials. Finally, we conclude with an overview of the data infrastructure and software efforts designed to make ML capabilities readily available to the wider scientific community.