Research map / Graph learning / GNN architectures
Towards Understanding Generalization of Graph Neural Networks
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 |
|---|---|---|---|
| Use SGC (linear, fixed propagation) instead of GCN | GCN-style node classifier · Cora, Citeseer, Pubmed | accuracy and loss gap: Cora acc gap 9.76 -> 5.33; Citeseer 22.11 -> 11.50 | Helped |
| Use APPNP (personalized PageRank propagation) instead of GCN | GCN-style node classifier · Cora, Citeseer, Pubmed | loss gap: Cora loss gap 0.30 -> 0.16; Citeseer 0.77 -> 0.25 | Helped |
| Increase GCN depth from two to six layers (ablation) | GCN · Cora, Citeseer | accuracy gap: Cora 9.76 -> 13.45; Citeseer 22.11 -> 26.48 | Hurt |
| Increase GCNII depth from two to six layers (ablation) | GCNII · Cora, Citeseer | accuracy gap: Cora 7.69 -> 6.24; Citeseer 14.85 -> 13.49 | No clear effect |
| Use GCNII (initial residual and identity mapping) instead of GCN | GCN-style node classifier · Cora, Citeseer, Pubmed | accuracy gap: Cora 9.76 -> 7.69; Citeseer 22.11 -> 14.85 | Helped |
| Use GPR-GNN (learnable spectral coefficients) instead of APPNP (ablation) | APPNP · Cora, Citeseer, Pubmed | accuracy and loss gap: Cora acc gap 7.72 -> 8.90; Citeseer 9.99 -> 19.08 | Hurt |
| Use SGC instead of GCN | GCN-style node classifier · Citeseer | test accuracy: 71.78 -> 74.37 | Helped |
| Use SGC or APPNP instead of GCN | GCN-style node classifier · Cora | test accuracy: GCN 85.91 vs SGC 82.39 and APPNP 79.14 | Hurt |
| Train on gradient-based transductive SGD with more iterations T (generalization gap of GCN loss) (ablation) | GCN · Cora, Citeseer | loss gap: loss gap increases with iterations | Hurt |
Closest papers by meaning
- A Manifold Perspective on the Statistical Generalization of Graph Neural Networks (ICML 2025 · similarity 0.93)
- Generalization Error of Graph Neural Networks in the Mean-field Regime (ICML 2024 · similarity 0.90)
- Towards Bridging Generalization and Expressivity of Graph Neural Networks (ICLR 2025 · similarity 0.90)
- Covered Forest: Fine-grained generalization analysis of graph neural networks (ICML 2025 · similarity 0.90)
- On the Interplay between Graph Structure and Learning Algorithms in Graph Neural Networks (ICML 2025 · similarity 0.90)
- Generalization of Graph Neural Networks Is Robust to Model Mismatch (AAAI 2025 · 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.