Fig. 7: Test MAE changes after fusing DCN and LOPE strategies. | npj Computational Materials

Fig. 7: Test MAE changes after fusing DCN and LOPE strategies.

From: DenseGNN: universal and scalable deeper graph neural networks for high-performance property prediction in crystals and molecules

Fig. 7

This figure compares the test MAE changes for GNN models on six different property datasets in Matbench after fusing the DCN and LOPE strategies. a–f correspond to the test MAE results on Phonons, Perovskites, Jdft2d, Log gvrh, Dielectric, and Log kvrh datasets, respectively. The models include GIN from the computer domain; HamNet from the molecular domain; and SchNet from the materials domain. The MAE results show improvements across all datasets for all models after fusing DCN and LOPE strategies. Specifically, DCN showed greater enhancement than LOPE, particularly in models from the molecular and computer domains like HamNet and GIN.

Back to article page