Substructure Counting Benchmark
1.9MAE (Triangle)ShareGNN
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ShareGNNTraining setting=one separate model for each motif2026.05 | 1.9 | 3.2 | 2.4 | 3.6 | 5.3 | 5.1 | |
| ShareGNNTraining setting=single model jointly for all six motifs2026.05 | 3.3 | 9.1 | 6.3 | 5.6 | 5.3 | 7.8 | |
| GNN-SSWL+2026.05 | 6.4 | 6.7 | 7.8 | 7.9 | 10.8 | 15.4 | |
| SUN (EGO+)2026.05 | 7.9 | 8 | 6.4 | 10.5 | 17 | 55 | |
| PPGN2026.05 | 8.9 | 9.6 | 14.8 | 9 | 13.7 | 16.7 | |
| GNN-AK+2026.05 | 12.3 | 11.2 | 15 | 12.6 | 26.8 | 58.4 |