Implicit Properties: Exploring Materials Representations in Machine Learning for Inorganic Crystalline Solids

27^th February 2025, 14:00
Tonio Buonassisi / Raul Ortega-Ochoa
MIT, US / Technical University of Denmark

Abstract

Recent advancements in machine learning (ML) for materials have demonstrated that "simple" materials representations (e.g., the chemical formula alone without structural information) can sometimes achieve competitive property prediction performance in common-tasks. Our physics-based intuition would suggest that such representations are "incomplete", which indicates a gap in our understanding. This work proposes a tomographic interpretation of structure-property relations of materials to bridge that gap by defining what is a material representation, material properties, the material and the relationships between these three concepts using ideas from information theory. We verify this framework performing an exhaustive comparison of property-augmented representations on a range of material's property prediction objectives, providing insight into how different properties can encode complementary information.