Research map / Time series / Long-term forecasting
Learning to Reconstruct Missing Data from Spatiotemporal Graphs with Sparse Observations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| add spatiotemporal cross-attention to SPIN-H (ablation) | SPIN-H · AQI-36 | MAE: 15.47 -> 10.89 | Helped |
| add temporal self-attention to SPIN-H (ablation) | SPIN-H · AQI-36 | MAE: 13.76 -> 10.89 | Helped |
| add spatiotemporal cross-attention to SPIN-H (ablation) | SPIN-H · METR-LA point missing | MAE: 2.18 -> 1.96 | Helped |
| add temporal self-attention to SPIN-H (ablation) | SPIN-H · METR-LA point missing | MAE: 2.21 -> 1.96 | Helped |
| sparse spatiotemporal attention (SPIN) instead of autoregressive GRIN | SPIN · PEMS-BAY block missing | MAE: 1.14 -> 1.06 | Helped |
| SPIN instead of GRIN in point-missing | SPIN · PEMS-BAY point missing | MAE: 0.67 -> 0.70 | Hurt |
| SPIN instead of GRIN at 95% missing rate | SPIN · METR-LA point missing 95% | MAE: 4.08 -> 2.89 | Helped |
| SPIN instead of GRIN at 95% missing rate | SPIN · AQI point missing 95% | MAE: 20.38 -> 15.51 | Helped |
| SPIN instead of spatiotemporal Transformer with mask token | SPIN · AQI-36 | MAE: 11.98 -> 11.77 | No clear effect |
| hierarchical attention (SPIN-H) instead of SPIN | SPIN-H · AQI-36 | MAE: 11.77 -> 10.89 | Mixed |
Closest papers by meaning
- Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling (ICML 2024 · similarity 0.90)
- Learning from Highly Sparse Spatio-temporal Data (NeurIPS 2024 · similarity 0.89)
- Motif-aware Graph Neural Networks for Networked Time Series Imputation (AAAI 2025 · similarity 0.88)
- Fully-Connected Spatial-Temporal Graph for Multivariate Time-Series Data (AAAI 2024 · similarity 0.87)
- Graph Structure Learning for Spatial-Temporal Imputation: Adapting to Node and Feature Scales (AAAI 2025 · similarity 0.86)
- Over-squashing in Spatiotemporal Graph Neural Networks (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.