Work

Understanding and mitigating distribution shifts for universal machine learning interatomic potentials

Year: 2025

Type: article

Abstract: We find common distribution shifts that pose challenges for universal machine learning interatomic potentials (MLIPs). We develop test-time refinement strategies that mitigate the shifts and provide insights into why MLIPs struggle to generalize.

Source: Digital Discovery

DOI: 10.1039/d5dd00260e

Authors: Tobias Kreiman, Aditi S. Krishnapriyan

Institutions: University of California, Berkeley, Lawrence Berkeley National Laboratory

Language: English

Topic: Machine Learning in Materials Science

Subfield: Materials Chemistry

Field: Materials Science

Domain: Physical Sciences

Open Access status: diamond

Funders: Toyota Research Institute, National Energy Research Scientific Computing Center, Office of Science, Lawrence Berkeley National Laboratory