Research map / Graph learning / GNN architectures
Improving Graph Neural Network Expressivity via Subgraph Isomorphism Counting
Giorgos Bouritsas, Fabrizio Frasca, Stefanos P. Zafeiriou, Michael M. Bronstein
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add message passing on top of structural identifiers (DeepSets to GSN) (ablation) | GSN · ZINC | MAE: 0.288 -> 0.108 | Improved |
| add message passing on top of structural identifiers (DeepSets to GSN) (ablation) | GSN · Collab | accuracy: 80.9 -> 85.5 | Improved |
| add message passing on top of structural identifiers (DeepSets to GSN) (ablation) | GSN · NCI1 | accuracy: 80.3 -> 83.5 | Improved |
| add message passing on top of structural identifiers (DeepSets to GSN) (ablation) | GSN · MUTAG | accuracy: 93.3 -> 92.8 | No clear effect |
| add message passing on top of structural identifiers (DeepSets to GSN) (ablation) | GSN · ogbg-molhiv | ROC-AUC: 77.34 -> 77.99 | Improved |
| add substructure counts to MPNN | MPNN · ZINC | MAE: 0.254 -> 0.140 | Improved |
| add edge-wise structural features and edge features | GSN · ZINC | MAE: 0.140 -> 0.115 | Improved |
| add random features as identifiers (MPNN-r) | MPNN · ZINC | MAE: 0.254 -> 0.322 | Hurt |
| increase GSN size to about 500K parameters (ablation) | GSN · ZINC | MAE: 0.140 -> 0.101 | Improved |
| add substructure vector fields to DGN instead of eigenvectors | DGN · ogbg-molhiv | test ROC-AUC: 0.7970 -> 0.8039 | Improved |
| add substructure counts to GIN+VN | GIN+VN · ogbg-molhiv | test ROC-AUC: 0.7707 -> 0.7799 | Improved |
| add substructure counts to GIN | GIN · Collab | accuracy: 80.2 -> 85.5 | Improved |
Closest papers by meaning
- A Flexible, Equivariant Framework for Subgraph GNNs via Graph Products and Graph Coarsening (NeurIPS 2024 · similarity 0.91)
- Homomorphism Counts for Graph Neural Networks: All About That Basis (ICML 2024 · similarity 0.91)
- Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness (ICLR 2024 · similarity 0.91)
- Homomorphism Expressivity of Spectral Invariant Graph Neural Networks (ICLR 2025 · similarity 0.90)
- Deep Homomorphism Networks (NeurIPS 2024 · similarity 0.90)
- Subgraphormer: Unifying Subgraph GNNs and Graph Transformers via Graph Products (ICML 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
