Meta AI Introduces UMA (Universal Models for Atoms): A Family of Universal Models for Atoms

Meta AI has introduced UMA (Universal Models for Atoms), a family of universal machine learning interatomic potentials (MLIPs) designed to approximate the accuracy of Density Functional Theory (DFT) while drastically reducing computational costs, achieving inference times of less than a second compared to hours for traditional DFT calculations. These models leverage scaling relations inspired by large language models (LLMs) to optimize the balance between dataset size, model complexity, and computational efficiency, addressing the longstanding challenge of creating MLIPs that generalize across diverse chemical tasks. By training on extensive datasets such as Alexandria and OMat24, UMA