Research map / Graph learning / GNN architectures
Simple and Deep Graph Convolutional 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 a deep GCNII with initial residual and identity mapping instead of the strongest reported baseline | GCN on Cora · Cora (semi-supervised) | accuracy: 83.5 -> 85.5 ± 0.5 | Helped |
| Use a deep GCNII with initial residual and identity mapping instead of the strongest reported baseline | GCN on Citeseer · Citeseer (semi-supervised) | accuracy: 72.7 -> 73.4 ± 0.6 | Helped |
| Use GCNII* rather than the strongest baseline | Node classification · Pubmed (semi-supervised) | accuracy: 80.1 -> 80.3 ± 0.4 | Helped |
| Use GCNII* rather than the strongest baseline | Inductive node classification · PPI | Micro-averaged F1: 99.36 -> 99.56 ± 0.02 | Helped |
| Add the separate-weight GCNII* variant | GCNII · Wisconsin (full-supervised) | accuracy: 74.12 (16) -> 81.57 (16) | Helped |
| Add the separate-weight GCNII* variant | GCNII · Citeseer (full-supervised) | accuracy: 77.08 (64) -> 77.13 (64) | Mixed |
| Increase depth while retaining initial residual and identity mapping | GCNII · Cora (semi-supervised) | accuracy: 82.2 -> 85.5 | Helped |
| Increase depth while retaining initial residual and identity mapping | GCNII · Pubmed (semi-supervised) | accuracy: 80.2 (16 layers) -> 79.7 (64 layers) | Mixed |
| Add identity mapping to vanilla GCN (ablation) | Vanilla GCN · Semi-supervised node classification | accuracy | No clear effect |
| Add an initial residual connection from the input layer (ablation) | Vanilla GCN · Semi-supervised node classification | accuracy | Helped |
| Add identity mapping alongside the initial residual connection (ablation) | GCN with initial residual · Semi-supervised node classification | accuracy | Helped |
| Use initial residual and identity mapping together rather than either technique alone (ablation) | GCNII · Semi-supervised node classification | accuracy | Helped |
Closest papers by meaning
- Simplifying Graph Convolutional Networks (ICML 2019 · similarity 0.95)
- Deeper Insights Into Graph Convolutional Networks for Semi-Supervised Learning (AAAI 2018 · similarity 0.93)
- L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks (CVPR 2020 · similarity 0.93)
- On Provable Benefits of Depth in Training Graph Convolutional Networks (NeurIPS 2021 · similarity 0.93)
- Graph Convolutional Kernel Machine versus Graph Convolutional Networks (NeurIPS 2023 · similarity 0.91)
- Break the Ceiling: Stronger Multi-scale Deep Graph Convolutional Networks (NeurIPS 2019 · 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.