When more data hurts: Optimizing data coverage while mitigating diversity-induced underfitting in an ultrafast machine-learned potential
Source abstract
Abstract
Machine-learned interatomic potentials (MLIPs) are becoming an essential tool in materials modeling. However, optimizing the generation of training data used to parametrize the MLIPs remains a significant challenge. This is because MLIPs can fail when encountering local enviroments too different from those present in the training data. The difficulty of determining a priori the environments that will be encountered during molecular dynamics simulation necessitates diverse, high-quality training data. This study investigates how training data diversity affects the performance of MLIPs using the Ultra-Fast force field (${\mathrm{UF}}3$) to model amorphous silicon nitride. We employ expert and autonomously generated data to create the training data and fit four force field variants to subsets of the data. Our findings reveal a critical balance in training data diversity: insufficient diversity hinders generalization, while excessive diversity can exceed the MLIP's learning capacity, reducing simulation accuracy. Specifically, we found that the ${\mathrm{UF}}3$ variant trained on a subset of the training data, in which nitrogen-rich structures were removed, offered vastly better prediction and simulation accuracy than any other variant. By comparing these ${\mathrm{UF}}3$ variants, we highlight the nuanced requirements for creating accurate MLIPs, emphasizing the importance of application-specific training data to achieve optimal performance in modeling complex material behaviors.
Source: openalex_inverted_index · Quality: baseline
Research context
OpenAlex topics: Machine Learning in Materials Science · Advanced Memory and Neural Computing · Ferroelectric and Negative Capacitance Devices
SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.
Affiliations
University of Florida
Sandia National Laboratories
Funding
National Science Foundation · Laboratory Directed Research and Development · Sandia National Laboratories · National Nuclear Security Administration
References (43)
Show source-provided references
1. 10.1063/1.46959710.1063/1.469597 ↗
2. 10.1021/jp972209d10.1021/jp972209d ↗
3. 10.1002/adma.20190276510.1002/adma.201902765 ↗
4. 10.1016/j.commatsci.2017.08.03110.1016/j.commatsci.2017.08.031 ↗
5. 10.1038/s41467-020-20427-210.1038/s41467-020-20427-2 ↗
6. 10.1038/s41597-020-0473-z10.1038/s41597-020-0473-z ↗
7. 10.1016/j.jcp.2014.12.01810.1016/j.jcp.2014.12.018 ↗
8. 10.1063/1.501764110.1063/1.5017641 ↗
9. 10.1103/PhysRevB.99.01410410.1103/PhysRevB.99.014104 ↗
10. 10.1080/00268976.2021.197642610.1080/00268976.2021.1976426 ↗
11. 10.1557/s43578-023-01239-810.1557/s43578-023-01239-8 ↗
12. 10.1021/acs.jpca.9b0872310.1021/acs.jpca.9b08723 ↗
13. 10.1557/s43578-023-01194-410.1557/s43578-023-01194-4 ↗
14. 10.1063/5.000508410.1063/5.0005084 ↗
15. 10.1103/PhysRevB.58.253910.1103/PhysRevB.58.2539 ↗
16. 10.1063/5.001305910.1063/5.0013059 ↗
17. 10.1038/s41524-022-00872-x10.1038/s41524-022-00872-x ↗
18. 10.1063/5.014675310.1063/5.0146753 ↗
19. 10.1016/0893-6080(89)90020-810.1016/0893-6080(89)90020-8 ↗
20. 10.1038/s41524-022-00721-x10.1038/s41524-022-00721-x ↗
21. 10.1103/PhysRevB.97.05430310.1103/PhysRevB.97.054303 ↗
22. 10.1038/s41524-023-01092-710.1038/s41524-023-01092-7 ↗
23. 10.1103/PhysRevB.62.311710.1103/PhysRevB.62.3117 ↗
24. 10.1016/j.commatsci.2017.12.00810.1016/j.commatsci.2017.12.008 ↗
25. M. Thorpe · 199710.1007/978-94-015-8832-4 ↗
26. 10.1103/PhysRevB.111.02410610.1103/PhysRevB.111.024106 ↗
27. 10.3390/nano1308135210.3390/nano13081352 ↗
28. 10.1021/acs.jpcc.4c0220510.1021/acs.jpcc.4c02205 ↗
29. 10.1557/s43578-023-01217-010.1557/s43578-023-01217-0 ↗
30. 10.1016/j.cpc.2021.10817110.1016/j.cpc.2021.108171 ↗
31. 10.1103/PhysRevLett.104.13640310.1103/PhysRevLett.104.136403 ↗
32. B. C. Revard · 2014
33. 10.1088/0953-8984/25/49/49540110.1088/0953-8984/25/49/495401 ↗
34. 10.1103/PhysRevB.93.05411710.1103/PhysRevB.93.054117 ↗
35. 10.1103/PhysRevB.47.55810.1103/PhysRevB.47.558 ↗
36. 10.1103/PhysRevB.49.1425110.1103/PhysRevB.49.14251 ↗
37. 10.1016/0927-0256(96)00008-010.1016/0927-0256(96)00008-0 ↗
38. 10.1103/PhysRevB.54.1116910.1103/PhysRevB.54.11169 ↗
39. 10.1103/PhysRevB.50.1795310.1103/PhysRevB.50.17953 ↗
40. 10.1103/PhysRevLett.77.386510.1103/PhysRevLett.77.3865 ↗
41. 10.1016/j.commatsci.2018.07.04310.1016/j.commatsci.2018.07.043 ↗
42. I. Batatia · 2022
43. 10.1038/s43588-022-00349-310.1038/s43588-022-00349-3 ↗