Research map / Graph learning / GNN architectures
Dissecting the Diffusion Process in Linear 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 |
|---|---|---|---|
| decouple terminal time T from propagation steps K (DGC instead of SGC) | SGC · Cora semi-supervised | test accuracy (%): 81.0 -> 83.3 | Helped |
| decouple T and K | SGC · Citeseer semi-supervised | test accuracy (%): 71.9 -> 73.3 | Helped |
| decouple T and K | SGC · Pubmed fully-supervised | test accuracy (%): 83.3 -> 89.4 | Helped |
| decouple T and K | SGC · Reddit | test accuracy (%): 94.9 -> 95.8 | Helped |
| compare linear DGC with non-linear APPNP | APPNP · Cora fully-supervised | test accuracy (%): 90.2 -> 88.2 | Hurt |
| increase propagation steps K at fixed terminal time T (ablation) | DGC · Cora | test accuracy | Helped |
| use 4th-order Runge-Kutta instead of Euler (ablation) | DGC · Cora | test accuracy | No clear effect |
| use self-loop Laplacian I-S instead of I-S_sym (ablation) | DGC · Pubmed | test accuracy (%): 79.8 -> 80.3 | Helped |
| use self-loop Laplacian I-S instead of I-S_sym (ablation) | DGC · Reddit | test accuracy (%): 95.8 -> 95.5 | Hurt |
| increase K from 2 to 100 (ablation) | DGC · Pubmed | total time: 65.3 ms -> 225.0 ms | Hurt |
| choose terminal time too small (T=1) or too large (T=10) rather than T=5.3 (ablation) | DGC · Cora | test accuracy | Hurt |
Closest papers by meaning
- Simplifying Graph Convolutional Networks (ICML 2019 · similarity 0.93)
- Diffusion Improves Graph Learning (NeurIPS 2019 · similarity 0.91)
- Deeper Insights Into Graph Convolutional Networks for Semi-Supervised Learning (AAAI 2018 · similarity 0.91)
- Simple and Deep Graph Convolutional Networks (ICML 2020 · similarity 0.91)
- L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks (CVPR 2020 · similarity 0.91)
- G$^2$CN: Graph Gaussian Convolution Networks with Concentrated Graph Filters (ICML 2022 · 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.