Research map / Graph learning / GNN architectures
G$^2$CN: Graph Gaussian Convolution Networks with Concentrated Graph Filters
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 G2CN (two Gaussian filters, b=0 and b=2) instead of DGC | linear GCN propagation · Cora / CiteSeer / PubMed (semi-supervised) | test accuracy (%): Cora 82.7 vs DGC 83.3; CiteSeer 73.8 vs 73.3; PubMed 80.4 vs 80.3 | Mixed |
| Use G2CN instead of DGC and BernNet | linear GCN propagation · Computers / Photo (fully supervised) | test accuracy (%): Computers 91.46 vs DGC 89.54; Photo 95.29 vs DGC 94.89 | Helped |
| Use G2CN instead of DGC | linear GCN propagation · Chameleon / Squirrel (heterophily) | test accuracy (%): Chameleon 73.61 vs 67.08; Squirrel 66.91 vs 50.51 | Helped |
| Use G2CN instead of DGC | linear GCN propagation · Texas / Cornell / Actor (heterophily) | test accuracy (%): Texas 96.72 vs 94.74; Cornell 94.11 vs 92.45; Actor 41.44 vs 40.35 (BernNet 41.79 higher on Actor) | Mixed |
| Use G2CN+MLP instead of DGC+MLP | linear GCN propagation with MLP head · OGB-Arxiv | test accuracy (%): 71.6 vs 71.2 | Helped |
| Use G2CN-Euler propagation with two Gaussian filters versus DGC for computation time | linear GCN training time · Pubmed / Computers / Photo (1000 epochs) | total training time (s): Pubmed 1.24s vs 1.04s; Computers 2.27s vs 2.01s; Photo 1.56s vs 1.40s | Hurt |
| Add Gaussian propagation with learnable Gaussian bases (G2CN) vs BernNet | BernNet-style propagation · Chameleon / Squirrel / Cornell / Texas (heterophily) | test accuracy (%): Chameleon 73.61 vs 54.11; Squirrel 66.91 vs 41.35; Cornell 94.11 vs 92.29 | Helped |
| Replace BernNet with G2CN on homophilic Computers/Photo | polynomial-basis propagation · Computers / Photo (fully supervised) | test accuracy (%): Computers 91.46 vs 87.64; Photo 95.29 vs 94.50 | Helped |
Closest papers by meaning
- Dissecting the Diffusion Process in Linear Graph Convolutional Networks (NeurIPS 2021 · similarity 0.91)
- Deeper Insights Into Graph Convolutional Networks for Semi-Supervised Learning (AAAI 2018 · similarity 0.90)
- Simplifying Graph Convolutional Networks (ICML 2019 · similarity 0.90)
- Simple Spectral Graph Convolution (ICLR 2021 · similarity 0.90)
- L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks (CVPR 2020 · similarity 0.90)
- HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.89)
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.