Research map / Graph learning / GNN architectures
DeloopSGNN: Revisiting Spectral GNNs Through the Lens of Spatial Aggregation
Duanyu Li, Huijun Wu, Min Xie, Kai Lu, Wenzhe Zhang, Zhenwei Wu, Yong Dong, Ruibo Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| replace A^k aggregation with acyclic-path aggregation (DeloopSGNN) vs BernNet | spectral GNN · Squirrel | accuracy: 48.3 ± 1.2 -> 56.4 ± 1.4 | Improved |
| use DeloopSGNN vs JacobiConv | spectral GNN · Squirrel | accuracy: 54.6 ± 1.6 -> 56.4 ± 1.4 | Improved |
| use DeloopSGNN vs GCN | spatial GNN · Squirrel | accuracy: 46.9 ± 0.6 -> 56.4 ± 1.4 | Improved |
| use DeloopSGNN vs GPRGNN | spectral GNN · Chameleon | accuracy: 52.9 ± 1.1 -> 69.3 ± 0.9 | Improved |
| use DeloopSGNN vs BernNet under Metattack 10% vs clean on Photo | robustness · Photo | accuracy: 94.8 ± 0.1 -> 92.9 ± 0.7 | Improved |
| use acyclic path constraint under strong Metattack 15% | DeloopSGNN vs GCN · Photo (Metattack 15%) | accuracy: 76.3 ± 2.7 -> 89.1 ± 1.8 | Improved |
| use acyclic path constraint under strong Metattack 15% | DeloopSGNN vs GCNSVD · Citeseer (Metattack 15%) | accuracy: 65.6 ± 1.4 -> 64.1 ± 1.1 | Hurt |
| use acyclic path constraint under PGD 15% | DeloopSGNN vs best baseline · Computers (PGD 15%) | accuracy: 73.2 ± 5.4 -> 79.0 ± 0.8 | Improved |
| use DeloopSGNN on clean graphs vs GCN-SVD | GCNSVD · Computers (clean) | accuracy: 77.0 ± 1.7 -> 90.7 ± 0.4 | Improved |
| use DeloopSGNN vs EvenNet | spectral GNN · Texas | accuracy: 93.8 ± 1.8 -> 93.0 ± 0.8 | No clear effect |
| use DeloopSGNN vs H2GCN | spatial GNN · Reed98 | accuracy: 69.2 ± 2.3 -> 75.1 ± 2.9 | Improved |
| use DeloopSGNN vs GAT | spatial GNN · Reed98 | accuracy: 76.8 ± 0.5 -> 75.1 ± 2.9 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- BernNet — BernNet: Learning Arbitrary Graph Spectral Filters via Bernstein Approximation (NeurIPS 2021) · better on 3 of 3 shared results
- EvenNet — EvenNet: Ignoring Odd-Hop Neighbors Improves Robustness of Graph Neural Networks (NeurIPS 2022) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.94)
- SLOG: An Inductive Spectral Graph Neural Network Beyond Polynomial Filter (ICML 2024 · similarity 0.92)
- On the Expressive Power of Spectral Invariant Graph Neural Networks (ICML 2024 · similarity 0.90)
- Spatio-Spectral Graph Neural Networks (NeurIPS 2024 · similarity 0.90)
- Position: Spectral GNNs Rely Less on Graph Fourier Basis than Conceived (ICML 2025 · similarity 0.90)
- AutoSGNN: Automatic Propagation Mechanism Discovery for Spectral Graph Neural Networks (AAAI 2025 · similarity 0.90)
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.
