Compositionally restricted attention-based network for materials property predictions

In this paper, we demonstrate an application of the Transformer self-attention mechanism in the context of materials science. Our network, the Compositionally Restricted Attention-Based network (CrabNet), explores the area of structure-agnostic materials property predictions when only a chemical formula is provided. Our results show that CrabNet's performance matches or exceeds current best-practice methods on nearly all of 28 total benchmark datasets. We also demonstrate how CrabNet's architecture lends itself towards model interpretability by showing different visualization approaches that are made possible by its design. We feel confident that CrabNet and its attention-based framework will be of keen interest to future materials informatics researchers.

Automated summary

In this paper, we demonstrated the application of the Transformer self-attention mechanism in materials science through CrabNet. CrabNet shows promising performance in structure-agnostic materials property predictions when only a chemical formula is provided, outperforming current best-practice methods on 28 benchmark datasets and offering model interpretability through its architecture. CrabNet and its attention-based framework are expected to spark interest among future materials informatics researchers.