Research map / Graph learning / GNN architectures
GraphNorm: A Principled Approach to Accelerating Graph Neural Network Training
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace InstanceNorm with GraphNorm (adds learnable shift alpha) (ablation) | GIN · ogbg-molhiv | ROC-AUC: +0.19 (77.73 vs 77.54) | No clear effect |
| Replace InstanceNorm with GraphNorm (adds learnable shift alpha) (ablation) | GCN · ogbg-molhiv | ROC-AUC: +0.12 (78.30 vs 78.18) | No clear effect |
| Replace InstanceNorm with GraphNorm (adds learnable shift alpha) (ablation) | GIN · IMDB-BINARY | test accuracy: +1.2 (76.0 vs 74.8) | Helped |
| Replace InstanceNorm with GraphNorm (adds learnable shift alpha) (ablation) | GIN · MUTAG | test accuracy: +1.1 (91.6 vs 90.5) | Helped |
| Replace InstanceNorm with GraphNorm (adds learnable shift alpha) (ablation) | GIN · PROTEINS | test accuracy: +0.9 (77.4 vs 76.5) | Helped |
| Replace InstanceNorm with GraphNorm (adds learnable shift alpha) (ablation) | GIN · NCI1 | test accuracy: +0.2 (81.4 vs 81.2) | Helped |
| Replace InstanceNorm with GraphNorm (adds learnable shift alpha) (ablation) | GIN · PTC | test accuracy: +0.2 (64.9 vs 64.7) | Helped |
| Replace InstanceNorm with GraphNorm (adds learnable shift alpha) (ablation) | GIN · RDT-B | test accuracy: +0.3 (93.5 vs 93.2) | Helped |
| Replace InstanceNorm with GraphNorm (adds learnable shift alpha) (ablation) | GIN · COLLAB | test accuracy: +0.2 (80.2 vs 80.0) | Helped |
| Use InstanceNorm (per-graph normalization) instead of BatchNorm or LayerNorm (ablation) | GCN/GIN · ogbg-molhiv | ROC-AUC: GCN: +1.96 over BatchNorm (78.18 vs 76.22); GIN: +0.93 (77.54 vs 76.61) | Helped |
| Adding a learnable-shift variant to BatchNorm (ablation) | BatchNorm · graph classification benchmarks | test accuracy | No clear effect |
| Use running statistics instead of batch-level statistics in BatchNorm (ablation) | BatchNorm · graph classification benchmarks | test accuracy | Hurt |
Closest papers by meaning
- Residual Connections and Normalization Can Provably Prevent Oversmoothing in GNNs (ICLR 2025 · similarity 0.89)
- Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification (NeurIPS 2024 · similarity 0.89)
- Graph Neural Networks Need Cluster-Normalize-Activate Modules (NeurIPS 2024 · similarity 0.88)
- A Manifold Perspective on the Statistical Generalization of Graph Neural Networks (ICML 2025 · similarity 0.88)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.88)
- Towards Scalable and Deep Graph Neural Networks via Noise Masking (AAAI 2025 · similarity 0.88)
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.