Research map / Graph learning / GNN architectures
Bag of Tricks for Training Deeper Graph Neural Networks: A Comprehensive Benchmark Study
Tianlong Chen, Kaixiong Zhou, Keyu Duan, Wenqing Zheng, Peihao Wang, Xia Hu, Zhangyang Wang
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 initial connection (skip) to 32-layer GCN (ablation) | GCN · Cora (32 layers) | test accuracy (%): 21.22 -> 78.74 | Improved |
| add residual connection to 32-layer GCN (ablation) | GCN · Cora (32 layers) | test accuracy (%): 21.22 -> 19.57 | No clear effect |
| add jumping connection to 32-layer GCN (ablation) | GCN · OGBN-ArXiv (32 layers) | test accuracy (%): 45.48 -> 71.87 | Improved |
| add dense connection to 32-layer GCN (ablation) | GCN · OGBN-ArXiv (32 layers) | test accuracy (%): 45.48 -> 70.94 | Improved |
| add identity mapping to 32-layer GCN (ablation) | GCN · Cora (32 layers) | test accuracy (%): 21.22 -> 40.57 | Improved |
| add identity mapping to 32-layer GCN (ablation) | GCN · PubMed (32 layers) | test accuracy (%): 38.77 -> 73.74 | Improved |
| add identity mapping to 2-layer GCN (ablation) | GCN · Citeseer (2 layers) | test accuracy (%): 71.46 -> 68.25 | Hurt |
| add GroupNorm to 32-layer SGC (ablation) | SGC · PubMed (32 layers) | test accuracy (%): 66.55 -> 74.62 | Improved |
| add NodeNorm to 32-layer GCN (ablation) | GCN · OGBN-ArXiv (32 layers) | test accuracy (%): 46.38 -> 29.94 | Hurt |
| add BatchNorm to 32-layer GCN (ablation) | GCN · OGBN-ArXiv (32 layers) | test accuracy (%): 46.38 -> 68.74 | Improved |
| add MeanNorm to deep GCN (ablation) | GCN · Cora (16 layers) | test accuracy (%): 21.78 -> 13.51 | Hurt |
| add DropEdge to 32-layer GCN (ablation) | GCN · OGBN-ArXiv (32 layers) | test accuracy (%): 41.96 -> 51.70 | Improved |
Closest papers by meaning
- L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks (CVPR 2020 · similarity 0.92)
- A Comprehensive Study on Large-Scale Graph Training: Benchmarking and Rethinking (NeurIPS 2022 · similarity 0.92)
- Fast and Deep Graph Neural Networks (AAAI 2020 · similarity 0.92)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · similarity 0.92)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.91)
- Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification (NeurIPS 2024 · similarity 0.91)
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.
