Case Studies in Representation Learning for Inverse Materials Design

Many grand challenges for the next generation of scientists and engineers will involve the design, manufacture, and maintenance of advanced materials, whose sought-after functions and properties will be derived from their yet-unknown internal structure. This relationship between structure and function is challenging to understand and even harder to predict because it is nonlinear, high dimensional, and results from physical phenomena at many scales. While traditional materials design has relied on human intuition to interpret patterns in known structure-function pairs and infer new materials with similar properties, modern data science approaches provide more efficient routes to “inverse design” of properties via structure.<br/>We have recently explored a variety of representation learning approaches in pursuit of a general strategy for materials design through deep generative modeling. I will describe three case studies where these approaches are applied to material design problems: High-Entropy Alloy chemistry designed for tailored mechanical properties, 3D-printed composite materials designed for optimal compliance, and macromolecules designed for optimal self-assembly behaviors. These three problems represent a broad range of materials topics with different physics, length scales, and traditional design strategies, demonstrating the transferability of the approach to arbitrary problems in materials design.