Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence

Benchmark Model Rank Results
graph-classification-on-cifar10-100kGatedGCN+#2Accuracy (%): 77.218 ± 0.381
graph-classification-on-malnet-tinyGatedGCN+#2Accuracy: 94.600±0.570
graph-classification-on-mnistGatedGCN+#4Accuracy: 98.712 ± 0.137
graph-classification-on-mnistGCN+#7Accuracy: 98.382 ± 0.095
graph-classification-on-peptides-funcGCN+#7AP: 0.7261 ± 0.0067
graph-property-prediction-on-ogbg-code2GatedGCN+#4Test F1 score: 0.1896 ± 0.0024Validation F1 score: 0.1742 ± 0.0027
graph-property-prediction-on-ogbg-molhivGatedGCN+#11Test ROC-AUC: 0.8040 ± 0.0164Ext. data: No
graph-property-prediction-on-ogbg-molpcbaGatedGCN+#9Test AP: 0.2981 ± 0.0024Ext. data: NoValidation AP: 0.3011 ± 0.0037
graph-property-prediction-on-ogbg-ppaGatedGCN+#1Ext. data: NoTest Accuracy: 0.8258 ± 0.0055
graph-property-prediction-on-ogbg-ppaGIN+#2Ext. data: NoTest Accuracy: 0.8107 ± 0.0053
graph-property-prediction-on-ogbg-ppaGCN+#3Ext. data: NoTest Accuracy: 0.8077 ± 0.0041
graph-regression-on-peptides-structGCN+#3MAE: 0.2421 ± 0.0016
graph-regression-on-zinc-500kGIN+#6MAE: 0.065
node-classification-on-clusterGatedGCN+#3Accuracy: 79.128 ± 0.235
node-classification-on-coco-spGatedGCN+#3macro F1: 0.3802 ± 0.0015
node-classification-on-pascalvoc-sp-1GatedGCN+#5macro F1: 0.4263 ± 0.0057
node-classification-on-patternGatedGCN+#3Accuracy: 87.029 ± 0.037