Research map / Graph learning / GNN architectures
DiffGCN: Graph Convolutional Networks via Differential Operators and Algebraic Multigrid Pooling
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 |
|---|---|---|---|
| Add AMG pooling | DiffGCN classification · ModelNet40 | Overall Accuracy: 93.5 -> 93.9 | Helped |
| Add AMG pooling | DiffGCN segmentation · S3DIS | mIoU: 61.1 -> 61.4 | Helped |
| Add AMG pooling | DiffGCN part segmentation · ShapeNet Parts | Mean IoU: 86.4 -> 86.4 | No clear effect |
| Reduce neighbors from K=20 to K=10 (ablation) | DiffGCN with mass, gradient, and Laplacian · ModelNet40 | Overall Accuracy: 93.5 -> 92.7 | Hurt |
| Reduce neighbors from K=20 to K=5 (ablation) | DiffGCN with mass, gradient, and Laplacian · ModelNet40 | Overall Accuracy: 93.5 -> 92.1 | Hurt |
| Add pooling with K=10 neighbors (ablation) | DiffGCN with mass, gradient, and Laplacian · ModelNet40 | Overall Accuracy: 92.7 -> 93.1 | Helped |
| Add pooling with K=5 neighbors (ablation) | DiffGCN with mass, gradient, and Laplacian · ModelNet40 | Overall Accuracy: 92.1 -> 92.3 | Helped |
| Add gradient and Laplacian to the mass term (ablation) | DiffGCN · ModelNet40 | Overall Accuracy: 88.2 -> 92.7 | Helped |
| Add the Laplacian term alongside mass and gradient (ablation) | DiffGCN · ModelNet40 | Overall Accuracy: 91.0 -> 92.7 | Helped |
| Add the gradient term alongside mass and Laplacian (ablation) | DiffGCN · ModelNet40 | Overall Accuracy: 91.9 -> 92.7 | Helped |
| Use mass and Laplacian instead of mass and gradient (ablation) | DiffGCN · ModelNet40 | Overall Accuracy: 91.0 -> 91.9 | Helped |
Closest papers by meaning
- Rethinking pooling in graph neural networks (NeurIPS 2020 · similarity 0.90)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · similarity 0.88)
- pathGCN: Learning General Graph Spatial Operators from Paths (ICML 2022 · similarity 0.88)
- Geometry-Aware Edge Pooling for Graph Neural Networks (NeurIPS 2025 · similarity 0.88)
- Dissecting the Diffusion Process in Linear Graph Convolutional Networks (NeurIPS 2021 · similarity 0.88)
- Going Deep: Graph Convolutional Ladder-Shape Networks (AAAI 2020 · similarity 0.88)
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.