Research map / Graph learning / GNN architectures
Towards Deeper Graph Neural Networks with Differentiable Group Normalization
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 |
|---|---|---|---|
| apply DGN instead of no normalization at 15 layers | GCN · Cora | test accuracy (%): 18.1 -> 75.2 | Helped |
| apply DGN instead of batch norm at 15 layers | GAT · Cora | test accuracy (%): 33.1 -> 71.8 | Helped |
| apply DGN instead of no normalization at 2 layers | GCN · Cora | test accuracy (%): 82.2 -> 82.0 | No clear effect |
| apply DGN instead of no normalization at 60 SGC layers | SGC · Citeseer | test accuracy (%): 66.3 -> 63.4 | Hurt |
| apply DGN instead of batch norm at 30 layers | GCN · CoauthorCS | test accuracy (%): 84.7 -> 84.4 | No clear effect |
| apply DGN instead of batch norm, missing features | GCN · Cora missing features | best test accuracy (%): 71.8 -> 76.3 | Helped |
| apply DGN instead of batch norm, missing features | SGC · CoauthorCS missing features | best test accuracy (%): 84.4 -> 85.8 | Helped |
| use average improvement of DGN over no normalization, batch and pair norm | GCN/GAT/SGC · missing-features node classification | average improvement (%): 37.8%, 7.1%, 12.8% over NN, BN, PN | Helped |
| increase number of groups from 1 to 10 in DGN (ablation) | DGN on GCN, 20 layers · Cora missing features | test accuracy (%): 74.7 -> 76.3 | Helped |
| set lambda near zero (0.001) in DGN, weakening group normalization (ablation) | DGN on GCN, 20 layers · Cora missing features | test accuracy | Hurt |
| vary group number G (ablation) | DGN on GCN, 20 layers · Cora missing features | test accuracy | No clear effect |
Closest papers by meaning
- Measuring and Relieving the Over-Smoothing Problem for Graph Neural Networks from the Topological View (AAAI 2020 · similarity 0.90)
- Curriculum-Enhanced Residual Soft An-Isotropic Normalization for Over-Smoothness in Deep GNNs (AAAI 2024 · similarity 0.89)
- Residual Connections and Normalization Can Provably Prevent Oversmoothing in GNNs (ICLR 2025 · similarity 0.89)
- Deep Graph Neural Networks via Posteriori-Sampling-based Node-Adaptative Residual Module (NeurIPS 2024 · similarity 0.88)
- Integrating Co-Training with Edge Discrimination to Enhance Graph Neural Networks Under Heterophily (AAAI 2025 · similarity 0.87)
- Graph Neural Networks Need Cluster-Normalize-Activate Modules (NeurIPS 2024 · similarity 0.86)
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.