Research map / Time series / Long-term forecasting
Towards Expressive Spectral-Temporal Graph Neural Networks for Time Series Forecasting
Ming Jin, Guangsi Shi, Yuan-Fang Li, Bo Xiong, Tian Zhou, Flora Dilys Salim, Liang Zhao, Lingfei Wu, Qingsong Wen, Shirui Pan
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 Gegenbauer basis instead of Monomial basis (ablation) | TGGC · PeMS04 | MAE: 59.41 -> 18.77 | Improved |
| use Gegenbauer basis instead of Bernstein basis (ablation) | TGGC · PeMS04 | MAE: 55.17 -> 18.77 | Improved |
| use Gegenbauer basis instead of Chebyshev basis (ablation) | TGGC · PeMS03 | MAE: 13.56 -> 13.52 | No clear effect |
| use Jacobi basis instead of Gegenbauer basis (ablation) | TGGC · PeMS03 | MAE: 13.52 -> 13.14 | Improved |
| use distinct polynomial coefficients along D dimensions (ablation) | TGGC · PeMS03 | MAE: 14.07 -> 13.52 | Improved |
| use distinct temporal filters per variable (ablation) | TGGC · PeMS04 | MAE: 19.12 -> 18.77 | Improved |
| use orthogonal space projections instead of random transformations (ablation) | TGGC · PeMS08 | RMSE: 28.84 -> 22.73 | Improved |
| add coarse-grained temporal frequency-domain model (ablation) | TGGC · PeMS04 | MAE: 19.15 -> 18.77 | Improved |
| add fine-grained temporal frequency-domain model (ablation) | TGGC · PeMS08 | MAE: 14.88 -> 14.55 | Improved |
| add nonlinearities to TGGC (ablation) | TGGC-dagger · PeMS08 | RMSE: 24.93 -> 22.43 | Improved |
| add spectral attention (ablation) | TGGC-dagger · PeMS03 | MAE: 13.42 -> 13.39 | No clear effect |
| use TGGC instead of StemGNN | spectral-temporal GNN · PeMS07 | MAE: 2.14 -> 1.92 | Improved |
Closest papers by meaning
- Designing Specialized Two-Dimensional Graph Spectral Filters for Spatial-Temporal Graph Modeling (AAAI 2025 · similarity 0.89)
- Simple and Efficient Heterogeneous Temporal Graph Neural Network (NeurIPS 2025 · similarity 0.88)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.88)
- Expand and Compress: Exploring Tuning Principles for Continual Spatio-Temporal Graph Forecasting (ICLR 2025 · similarity 0.88)
- Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling (ICML 2024 · similarity 0.87)
- Temporal-Aware Evaluation and Learning for Temporal Graph Neural Networks (AAAI 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.
