Fig. 4: AdsMT performance on global minimum adsorption energy (GMAE) prediction and transfer learning.
From: A multi-modal transformer for predicting global minimum adsorption energy
a The prediction success rate (SR) and mean absolute error (MAE) of AdsMT models with different graph encoders35,46,47,48,49,50 on the three GMAE datasets. b Schematic illustration of our transfer learning strategy. Each AdsMT model is pre-trained on the OC20-local minimum adsorption energy (LMAE) dataset, and then fine-tuned on a GMAE dataset with the graph encoder parameters selectively frozen. c AdsMT's performance gains after transfer learning using different graph encoders48,49,50.
