The discovery of new materials is essential for enabling technological advancements. Computational approaches seek to effectively learn the manifold of stable crystal structures within an infinite design space, and also rely on robust benchmarks and minimal, information-rich datasets for meaningful evaluation. In this talk, I will introduce Open Materials Generation (OMatG), a unifying framework for the generative design and discovery of inorganic crystalline materials using stochastic interpolants. The performance and benchmarks / datasets for materials generation will be discussed in the context of two tasks: crystal structure prediction (CSP) for given compositions and de novo generation (DNG) aimed at discovering entirely novel, stable, and unique structures. By revising and extending common metrics and datasets, we demonstrate OMatG’s state-of-the-art performance, surpassing purely flow- and diffusion-based implementations. These results underscore the importance of flexible deep learning frameworks and sensible benchmarks to accelerate progress in materials science.
Copyright. All Rights Reserved.