Research map / Graph learning / Graph self-supervised learning
How Framelets Enhance Graph Neural Networks
Xuebin Zheng, Bingxin Zhou, Junbin Gao, Yuguang Wang, Pietro Lió, Ming Li, Guido Montufar
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 |
|---|---|---|---|
| Use shrinkage instead of ReLU activation (ablation) | UFGConv · Cora | test accuracy: 83.6 ±0.6 -> 83.0 ±0.5 | Hurt |
| Use shrinkage instead of ReLU activation (ablation) | UFGConv · Citeseer | test accuracy: 72.7 ±0.6 -> 71.0 ±0.6 | Hurt |
| Use shrinkage instead of ReLU activation (ablation) | UFGConv · Pubmed | test accuracy: 79.9 ±0.1 -> 79.4 ±0.4 | Hurt |
| Use shrinkage instead of ReLU activation (ablation) | UFGConv · ogbn-arxiv | test accuracy: 71.97 ±0.12 -> 70.04 ±0.22 | Hurt |
| Aggregate framelet coefficients by power spectrum rather than sum (ablation) | UFGPool · Mutagenicity | test accuracy: 81.59 ±1.40 -> 82.05 ±1.28 | Improved |
| Aggregate framelet coefficients by power spectrum rather than sum (ablation) | UFGPool · QM7 | MSE: 41.74 ±0.84 -> 41.67 ±0.95 | Improved |
| Aggregate framelet coefficients by power spectrum rather than sum (ablation) | UFGPool · PROTEINS | test accuracy: 77.77 ±2.60 -> 77.23 ±2.40 | Hurt |
| Aggregate framelet coefficients by power spectrum rather than sum (ablation) | UFGPool · D&D | test accuracy: 80.92 ±1.68 -> 79.83 ±1.88 | Hurt |
| Aggregate framelet coefficients by power spectrum rather than sum (ablation) | UFGPool · NCI1 | test accuracy: 77.88 ±1.24 -> 77.54 ±2.24 | Hurt |
| Aggregate framelet coefficients by power spectrum rather than sum (ablation) | UFGPool · ogbg-molhiv | ROC-AUC: 78.80 ±0.56 -> 78.36 ±0.77 | Hurt |
| Vary framelet dilation (ablation) | UFGConv · Cora, Citeseer, and Pubmed | test accuracy | No clear effect |
| Use scale level 1 instead of the tested higher scale levels (ablation) | UFGConv · Cora, Citeseer, and Pubmed | test accuracy | Hurt |
Closest papers by meaning
- Deep Hypergraph Neural Networks with Tight Framelets (AAAI 2025 · similarity 0.88)
- SEA-GWNN: Simple and Effective Adaptive Graph Wavelet Neural Network (AAAI 2024 · similarity 0.88)
- A General Graph Spectral Wavelet Convolution via Chebyshev Order Decomposition (ICML 2025 · similarity 0.88)
- WaveNet: Tackling Non-stationary Graph Signals via Graph Spectral Wavelets (AAAI 2024 · similarity 0.87)
- ASWT-SGNN: Adaptive Spectral Wavelet Transform-Based Self-Supervised Graph Neural Network (AAAI 2024 · similarity 0.87)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · 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.
