Fig. 1: A schematic representation of the key concepts of the MACE-freeze method.
From: Fine-tuning foundation models of materials interatomic potentials with frozen transfer learning
a The workflow for training a model based on MACE-MP using transfer learning and using it to generate accurate data to fit a fast ACE surrogate model. b Transfer learning by freezing parameters in the model layers. c The atomic systems used in our benchmarks.
