An application of the Transformer self-attention mechanism in the field of materials science. The Compositionally Restricted Attention-Based network (CrabNet), explores the area of structure-agnostic materials property predictions when only a chemical formula is provided. The results show that CrabNet’s performance matches or exceeds current best-practice methods on nearly all of 28 total benchmark datasets. its architecture lends itself towards model interpretability by showing different visualization approaches that are made possible by its design.
Wang, Anthony Yu-Tung, et al. "Compositionally restricted attention-based network for materials property predictions." npj Computational Materials 7.1 (2021): 1-10.
Attention-based network, Materials property,
Curator: qinyang