Research map / Graph learning / Graph self-supervised learning
Spectral Invariant Learning for Dynamic Graphs under Distribution Shifts
Zeyang Zhang, Xin Wang, Ziwei Zhang, Zhou Qin, Weigao Wen, Hui Xue', Haoyang Li, Wenwu Zhu
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 |
|---|---|---|---|
| use SILD instead of DIDA | dynamic GNN OOD methods · Collab | AUC (%): 81.87 -> 84.09 | Improved |
| use SILD instead of DIDA | dynamic GNN OOD methods · Yelp | AUC (%): 75.92 -> 78.65 | Improved |
| use SILD instead of DIDA | dynamic GNN OOD methods · Aminer15 | accuracy (%): 50.34 -> 52.35 | Improved |
| use SILD instead of DIDA | dynamic GNN OOD methods · Aminer17 | accuracy (%): 44.69 -> 45.54 | Improved |
| use SILD instead of DIDA under strongest shift | dynamic GNN OOD methods · Link-Synthetic shift 0.8 | AUC (%): 72.59 -> 78.01 | Improved |
| use SILD instead of GroupDRO under strongest shift | dynamic GNN OOD methods · Node-Synthetic shift 0.8 | accuracy (%): 31.03 -> 38.64 | Improved |
| use SILD instead of DIDA at low shift | dynamic GNN OOD methods · Node-Synthetic shift 0.4 | accuracy (%): 43.33 -> 43.62 | No clear effect |
| add general OOD method GroupDRO on top of DySAT | DySAT · Yelp | AUC (%): 66.09 -> 66.97 | No clear effect |
| add general OOD method GroupDRO on top of DySAT | DySAT · Aminer15 | accuracy (%): 48.41 -> 48.73 | No clear effect |
| remove invariant spectral filtering (lambda=0) (ablation) | SILD · all datasets | performance | Improved |
| remove disentangled spectrum masks (ablation) | SILD · all datasets | performance | Improved |
| vary hyperparameter lambda too small or too large | SILD · most datasets | AUC | Hurt |
Closest papers by meaning
- HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.88)
- Beyond Spatio-Temporal Representations: Evolving Fourier Transform for Temporal Graphs (ICLR 2024 · similarity 0.88)
- Topology-Aware Dynamic Reweighting for Distribution Shifts on Graph (ICML 2025 · similarity 0.87)
- Position: Spectral GNNs Rely Less on Graph Fourier Basis than Conceived (ICML 2025 · similarity 0.87)
- Spectral Graph Neural Networks are Incomplete on Graphs with a Simple Spectrum (NeurIPS 2025 · similarity 0.87)
- Equivariant Machine Learning on Graphs with Nonlinear Spectral Filters (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.
