Improving Graph Neural Network Expressivity via Subgraph Isomorphism Counting

Benchmark Model Rank Results
graph-property-prediction-on-ogbg-molhivdirectional GSN#13Test ROC-AUC: 0.8039 ± 0.0090Ext. data: No
graph-property-prediction-on-ogbg-molhivGSN#25Test ROC-AUC: 0.7799 ± 0.0100Ext. data: No
graph-regression-on-zinc-100kGSN#2MAE: 0.115
graph-regression-on-zinc-500kGSN#20MAE: 0.101