Research map / Graph learning / GNN architectures
Binary Graph Neural Networks
Check 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 three-stage cascaded distillation instead of direct distillation or no distillation (ablation) | Binary DGCNN (BF1 and BF2) · ModelNet40 | accuracy (%): 61.26 -> 81.04; 62.84 -> 81.04 | Helped |
| Use Hamming k-NN with binary node features and real-valued weights (stage 2) | Binary DGCNN · ModelNet40 | accuracy (%): 91.57 vs 92.89; 90.32 vs 92.89 | Helped |
| Binarize both weights and node features rather than retaining real-valued weights (ablation) | Binary DGCNN · ModelNet40 | accuracy (%): 91.57 -> 81.08; 90.32 -> 81.32 | Hurt |
| Use PReLU instead of ReLU or no additional activation (ablation) | Binary DGCNN edge messages · ModelNet40 | accuracy (%): 76.22 -> 81.08; 76.09 -> 81.08 | Helped |
| Add mean- or median-centering at edge aggregation or global pooling (ablation) | Binary DGCNN · ModelNet40 | accuracy (%): 81.08 -> 51.05; 81.08 -> 75.93; 81.08 -> 67.87; 81.08 -> 60.82 | Hurt |
| Replace Gaussian RBF similarity with Hamming-based similarity for LSP (ablation) | Binary DGCNN · ModelNet40 | accuracy (%): 81.08 -> 9.40 | Hurt |
| Use LSP with Gaussian RBF similarity (ablation) | DGCNN binarization · ModelNet40 | accuracy (%): 81.04 -> 81.32; 81.00 -> 81.52; 91.05 -> 90.52 | Mixed |
| Use L2 rather than Hamming distance for k-NN (ablation) | Full-binary BF1/BF2 DGCNN · ModelNet40 | accuracy (%): 81.32 -> 81.00; 81.08 -> 81.52 | No clear effect |
| Use binary GraphSAGE instead of full-precision GraphSAGE | GraphSAGE · OGBN-Products | mean accuracy: 0.7862 -> 0.7300 | Hurt |
| Use logit matching to train binary GraphSAGE (ablation) | Binary GraphSAGE · OGBN-Products | mean accuracy: 0.7300 vs 0.7260 | No clear effect |
| Use binary GraphSAGE instead of full-precision GraphSAGE | GraphSAGE · OGBN-Proteins | mean accuracy: 0.7734 -> 0.7497 | Hurt |
| Binarize DGCNN and optimize Hamming-distance computation | DGCNN on Raspberry Pi 4B · Raspberry Pi 4B | speedup: 2x | Helped |
Closest papers by meaning
- Bi-GCN: Binary Graph Convolutional Network (CVPR 2021 · similarity 0.92)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · similarity 0.87)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.87)
- G-Net: A Provably Easy Construction of High-Accuracy Random Binary Neural Networks (NeurIPS 2025 · similarity 0.86)
- What do Graph Neural Networks learn? Insights from Tropical Geometry (NeurIPS 2024 · similarity 0.86)
- Training Binary Neural Networks via Gaussian Variational Inference and Low-Rank Semidefinite Programming (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.