Research map / Graph learning / GNN architectures
Convolutional Neural Networks on Graphs with Chebyshev Approximation, Revisited
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 |
|---|---|---|---|
| constrain Chebyshev coefficients by reparameterizing wk as wk/k (ablation) | ChebBase · Cora | accuracy: 79.29±0.36 -> 82.66±0.28 | Helped |
| constrain Chebyshev coefficients by reparameterizing wk as wk/k (ablation) | ChebBase · Pubmed | accuracy: 78.07±0.32 -> 79.25±0.31 | Helped |
| increase ChebNet polynomial order K from 2 to 10 (ablation) | ChebNet · Cora | accuracy: 80.54±0.38 -> 74.91±0.52 | Hurt |
| increase ChebNet polynomial order K from 2 to 10 (ablation) | ChebNet · Pubmed | accuracy: 75.52±0.75 -> 65.91±1.71 | Hurt |
| replace ChebNet's free coefficients with Chebyshev interpolation (ChebNetII) | ChebNet · Chameleon semi-supervised | accuracy: 37.15±1.49 -> 43.42±3.54 | Helped |
| replace ChebNet with ChebNetII | ChebNet · Squirrel full-supervised | accuracy: 40.81±0.42 -> 57.72±0.59 | Helped |
| use Chebyshev interpolation basis instead of Bernstein basis | spectral GNN · Squirrel full-supervised | accuracy: 51.39±0.92 -> 57.72±0.59 | Helped |
| use Chebyshev basis with free coefficients (ChebBase) instead of Monomial basis | GPR-GNN · Cora | accuracy: 83.95±0.22 -> 79.29±0.36 | Hurt |
| ChebNetII vs best scalable baseline on ogbn-papers100M | ogbn-papers100M · ogbn-papers100M | accuracy: 65.89±0.35 -> 67.18±0.32 | Helped |
| ChebNetII vs LINKX on arXiv-year | ChebNetII · arXiv-year | accuracy: 56.00±1.34 vs 48.53±0.31 | Hurt |
| ChebNetII vs GCNII on full-supervised Pubmed | ChebNetII · Pubmed full-supervised | accuracy: 88.93±0.29 vs 89.94±0.31 | No clear effect |
Closest papers by meaning
- Unifying Homophily and Heterophily for Spectral Graph Neural Networks via Triple Filter Ensembles (NeurIPS 2024 · similarity 0.91)
- A General Graph Spectral Wavelet Convolution via Chebyshev Order Decomposition (ICML 2025 · similarity 0.90)
- Return of ChebNet: Understanding and Improving an Overlooked GNN on Long Range Tasks (NeurIPS 2025 · similarity 0.90)
- HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.90)
- Position: Spectral GNNs Rely Less on Graph Fourier Basis than Conceived (ICML 2025 · similarity 0.88)
- SLOG: An Inductive Spectral Graph Neural Network Beyond Polynomial Filter (ICML 2024 · 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.