An entropy-maximized active learning method for MLIPs that selects diverse configurations using per-atom and per-config entropy measures, yielding 3-10x lower energy MAE than random MD sampling on carbon, silicon, and NaCl with 100-800 training structures.
Releases the Allegro-OAM-L foundation potential; pretrained on OMat24 and fine-tuned on MPtrj+sAlex
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cond-mat.mtrl-sci 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Dataset-aware entropy-maximized active learning for machine-learned interatomic potentials
An entropy-maximized active learning method for MLIPs that selects diverse configurations using per-atom and per-config entropy measures, yielding 3-10x lower energy MAE than random MD sampling on carbon, silicon, and NaCl with 100-800 training structures.