Research map / Graph learning / Knowledge and temporal graphs
A General Spatio-Temporal Backbone with Scalable Contextual Pattern Bank for Urban Continual Forecasting
Aoyu Liu, Yaying Zhang
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 STBP frequency-domain backbone plus contextual pattern bank instead of best baseline EAC | continual forecasting · PEMS-Stream | MAE: 15.67 -> 12.31 | Improved |
| use STBP instead of best baseline | continual forecasting · CA-Stream | MAE: 20.20 -> 15.77 | Improved |
| use STBP instead of best baseline | continual forecasting · AIR-Stream | MAE: 24.21 -> 23.64 | Improved |
| lightweight prompt-based adaptation on a frozen backbone rather than full fine-tuning | continual forecasting baselines · PEMS-Stream | MAE: 16.86 -> 15.67 | Improved |
| train with only 10% data in later periods (few-shot), STBP vs EAC | STBP · PEMS-Stream 10% | MAE: 16.13 -> 13.58 | Improved |
| STBP vs EAC under few-shot | STBP · CA-Stream 10% | MAE: 20.94 -> 17.11 | Improved |
| add contextual pattern bank (vs Retrain/Online without it) (ablation) | STBP · PEMS-Stream, CA-Stream, AIR-Stream | MAE, RMSE, MAPE | Improved |
| add the spatio-temporal backbone (FreNet+DLGA) instead of CNN+GCN (ablation) | STBP · three datasets | MAE, RMSE, MAPE | Improved |
| add DLGA dual-stream linear graph attention (ablation) | STBP backbone · three datasets | MAE, RMSE, MAPE | Improved |
| backbone alone with online training, no pattern bank, vs EAC (ablation) | STBP backbone · traffic datasets | MAE | No clear effect |
| increase feature dimension d from 64 to 256 (ablation) | STBP · PEMS-Stream, CA-Stream, AIR-Stream | MAE, RMSE | Mixed |
| use linear attention instead of full O(N^2) attention (ablation) | STBP · toy dataset | computational cost | Improved |
Closest papers by meaning
- Expand and Compress: Exploring Tuning Principles for Continual Spatio-Temporal Graph Forecasting (ICLR 2025 · similarity 0.92)
- Information Bottleneck-guided MLPs for Robust Spatial-temporal Forecasting (ICML 2025 · similarity 0.88)
- Towards Dynamic Spatial-Temporal Graph Learning: A Decoupled Perspective (AAAI 2024 · similarity 0.88)
- ST-FiT: Inductive Spatial-Temporal Forecasting with Limited Training Data (AAAI 2025 · similarity 0.88)
- STRAP: Spatio-Temporal Pattern Retrieval for Out-of-Distribution Generalization (NeurIPS 2025 · similarity 0.86)
- How Different from the Past? Spatio-Temporal Time Series Forecasting with Self-Supervised Deviation Learning (NeurIPS 2025 · similarity 0.86)
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.
