Research map / Probabilistic and scientific modeling / Molecules and proteins
Spatiotemporal Imputation with Graph-Informed Flow Matching
Zepeng Zhang, Aref Einizade, Jhony H. Giraldo, Olga Fink
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 |
|---|---|---|---|
| replace Gaussian prior with graph-informed prior (spatial + temporal filtering) (ablation) | GiFlow · Air-36 (point missing 20%) | MAE: 12.79 -> 9.54 | Improved |
| add spatial-only filtering prior vs Gaussian (ablation) | FM · Air-36 (point missing 20%) | MAE: 12.79 -> 9.75 | Improved |
| add temporal filtering on top of spatial-only prior (ablation) | GiFlow · Air-36 (point missing 20%) | MAE: 9.75 -> 9.54 | Improved |
| use graph-informed prior (ablation) | GiFlow · Air-36 (point missing 20%) | transport cost: 299.62 -> 104.29 | Improved |
| add spatial attention (ablation) | GiFlow · Air-36 (point missing 20%) | MAE: 10.05 -> 9.54 | Improved |
| add temporal attention (ablation) | GiFlow · Air-36 (point missing 20%) | MAE: 9.87 -> 9.54 | Improved |
| add spatiotemporal propagation (ablation) | GiFlow · Air-36 (point missing 20%) | MAE: 10.40 -> 9.54 | Improved |
| use both spatial and temporal attention (ablation) | GiFlow · Air-36 (point missing 20%) | MAE: 10.22 -> 9.54 | Improved |
| vary graph threshold from 0.1 to 0.02 (ablation) | GiFlow · Air-36 (point missing 20%) | MAE: 9.54 -> 10.33 | Hurt |
| vary graph threshold from 0.1 to 0.6 (ablation) | GiFlow · Air-36 (point missing 20%) | MAE: 9.54 -> 11.15 | Hurt |
| moderate graph threshold 0.05 to 0.4 (ablation) | GiFlow · Air-36 (point missing 20%) | MAE: 9.97 to 9.94 vs 9.54 at 0.1 | No clear effect |
| reduce Euler steps from 20 to 5 (ablation) | GiFlow · Air-36 (point missing 20%) | MAE: 9.54 -> 9.81 | No clear effect |
Closest papers by meaning
- RDPI: A Refine Diffusion Probability Generation Method for Spatiotemporal Data Imputation (AAAI 2025 · similarity 0.91)
- Graph Structure Learning for Spatial-Temporal Imputation: Adapting to Node and Feature Scales (AAAI 2025 · similarity 0.89)
- Learning from Highly Sparse Spatio-temporal Data (NeurIPS 2024 · similarity 0.88)
- Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling (ICML 2024 · similarity 0.85)
- MissPath-FM: Flow Matching with Structured Priors for Partially Observed Time Series (NeurIPS 2026 · similarity 0.85)
- Diffusion Transformers for Imputation: Statistical Efficiency and Uncertainty Quantification (NeurIPS 2025 · similarity 0.84)
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.
