Research map / Graph learning / GNN architectures
DFNets: Spectral CNNs for Graphs with Feedback-Looped 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 |
|---|---|---|---|
| add feedback-looped filters instead of Chebyshev filters (ablation) | spectral CNN without dense connections · Cora, Citeseer, and Pubmed | accuracy: +1.4%, +1.7%, and +7.3% | Helped |
| add feedback-looped filters instead of Cayley filters (ablation) | spectral CNN without dense connections · Cora | accuracy: +0.7% | Helped |
| add self-attention to DFNet | DFNet · Cora, Citeseer, Pubmed, and NELL | accuracy: +0.8, +0.5, +0.9, and +0.5 percentage points | Helped |
| add scaled-normalization and cut-off frequency together rather than using either technique alone (ablation) | DFNet, DFNet-ATT, and DF-ATT · Cora, Citeseer, and Pubmed | accuracy | Helped |
| use scaled-normalization (ablation) | feedback-looped filters · Cora, Citeseer, and Pubmed | accuracy | Helped |
| use cut-off frequency (ablation) | feedback-looped filters · Cora, Citeseer, and Pubmed | accuracy | Helped |
| set feedback and feedforward orders to p=5 and q=3 rather than other tested orders (ablation) | DFNet · Cora, Citeseer, and Pubmed | accuracy | Helped |
| set q larger than p (ablation) | DFNet feedback-looped filters · Cora, Citeseer, and Pubmed | accuracy | Hurt |
| use p that is too low or too high (ablation) | DFNet feedback-looped filters · Cora, Citeseer, and Pubmed | accuracy | Hurt |
| use DFNet instead of AdaLNet | node classification · Pubmed, 0.03% training split | accuracy: 59.3 ± 6.6 vs 61.0 ± 8.7 | Hurt |
Closest papers by meaning
- HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.89)
- SLOG: An Inductive Spectral Graph Neural Network Beyond Polynomial Filter (ICML 2024 · similarity 0.88)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.88)
- Spatio-Spectral Graph Neural Networks (NeurIPS 2024 · similarity 0.88)
- Unifying Homophily and Heterophily for Spectral Graph Neural Networks via Triple Filter Ensembles (NeurIPS 2024 · similarity 0.87)
- A Manifold Perspective on the Statistical Generalization of Graph Neural Networks (ICML 2025 · similarity 0.87)
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.