Robust and Interpretable Adaptation of Equivariant Materials Foundation Models via Sparsity-promoting Fine-tuning
Researchers have developed a novel fine-tuning method for E(3)-equivariant materials foundation models, which are used to approximate potential energy surfaces. This sparsity-promoting technique selectively updates a small fraction of model parameters, achieving performance comparable to or better than full fine-tuning on energy and force prediction tasks. The method also demonstrates effectiveness in specialized applications like magnetic moment prediction and offers interpretable insights into model behavior, such as highlighting d-orbital contributions in transition metal systems. AI
IMPACT This method could enable more efficient and interpretable adaptation of specialized AI models for materials science research.