Cui et al. present a test-time adaptation framework that refines machine learning models on-the-fly, reducing force and energy errors by ~30% and enabling stable molecular dynamics simulations without extra data. https://www.nature.com/articles/s41467-025-57101-4
Comments