SKM 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 31: Data Driven Materials Science: Big Data and Work Flows – Machine Learning
MM 31.2: Vortrag
Mittwoch, 29. März 2023, 16:00–16:15, SCH A 251
Critical Assessment of Uncertainty Estimates of Machine- Learning Potentials — •Shuaihua Lu1,2, Luca M. Ghiringhelli1, Christian Carbogno1, and Matthias Scheffler1 — 1Novel Ma- terials Discovery at the FHI of the Max-Planck-Gesellschaft and IRIS- Adlershof of the Humboldt-Universität zu Berlin, Berlin, Germany — 2School of Physics, Southeast University, Nanjing, China
Machine-learning potentials (MLP) trained on first-principles datasets are becoming increasingly popular since they enable the treatment of larger system sizes and longer time scales compared to straight ab initio techniques. A key aspect for the use of these MLPs is to reliably assess the accuracy viz. uncertainty of the predictions, e.g., by training an ensemble of models. Here, we critically examine the robustness of such uncertainty predictions using equivariant message-passing neural networks as an example [1]. We train an ensemble of models on liquid silicon simulated at the gradient-corrected density-functional-theory level and compare the predicted uncertainties with actual errors for various test sets, including liquid silicon at different temperatures and out-of-training-domain data such as solid phases with and without point defects as well as surfaces. These studies reveal that the predicted uncertainties are often overconfident. This is ascribed to the insufficient diversity in the members of the ensemble, as measured via error correlations. [1] S. Batzner et al., Nat. commun. 13, 2453 (2022).