Fig. 6: Figures of merit (FOM) for f4 models relative to the custom ACE trained on the Ti-Al-V data.
From: Fine-tuning foundation models of materials interatomic potentials with frozen transfer learning
Panel (a) shows how the FOMs of the transfer models (f4) improve as the size of the database increases. Panel (b) displays the FOMs of an ensemble of f4 10% models, quantifying the uncertainty obtained via random sampling of the training set.
