The field of materials science relies on the development of structure-processing-property relationships. However, as researchers study increasing complex and detailed phenomena, the relationships between structure, processing, and properties are rarely low-dimensional and become increasingly difficult for humans to observe, understand, and exploit. Machine learning offers the possibility of capturing trends in high dimensional data and could serve as a critical proxy for simulating materials properties.
In this talk, we will describe the unique aspects of materials informatics as they relate to material properties prediction. Specifically, we will cover materials property databases and repositories, composition and structure-based featurization, algorithm choice, extrapolation to new high-performing properties, and inverse design approaches.