Research map / Time series / Long-term forecasting
Filling the G_ap_s: Multivariate Time Series Imputation by Graph Neural Networks
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 |
|---|---|---|---|
| Use bidirectional graph recurrent imputation with spatial decoding instead of the strongest baseline | Time-series imputation · AQI and AQI-36, in-sample and out-of-sample | MAE: AQI-36 in-sample: 12.23 { $\pm$ 0.17} -> 10.51 { $\pm$ 0.28}; AQI-36 out-of-sample: 14.50 { $\pm$ 0.35} -> 12.08 { $\pm$ | Helped |
| Use bidirectional graph recurrent imputation with spatial decoding instead of the strongest baseline | Time-series imputation · PEMS-BAY, METR-LA, and CER-E, block and point missing | MAE: PEMS-BAY block: 1.59 { $\pm$ 0.00} -> 1.14 { $\pm$ 0.01}; PEMS-BAY point: 1.11 { $\pm$ 0.00} -> 0.67 { $\pm$ 0.00}; METR | Helped |
| Use graph-based GRIN rather than BRITS | Synthetic particle imputation · Fixed-charge and varying-charge particle systems | MSE: Fixed charge: 0.0878 { $\pm$ 0.0002} -> 0.0061 { $\pm$ 0.0010}; varying charge: 0.0840 { $\pm$ 0.0001} -> 0.0074 { $\pm$ | Helped |
| Add the spatial decoder (ablation) | GRIN · AQI, METR-LA block and point missing, CER-E block and point missing | MAE: AQI: 15.40 { $\pm$ 0.14} -> 14.73 { $\pm$ 0.15}; METR-LA block: 2.32 { $\pm$ 0.01} -> 2.03 { $\pm$ 0.00}; METR-LA point: | Mixed |
| Use the spatial decoder rather than a denoising decoder (ablation) | GRIN · AQI, METR-LA block and point missing, CER-E block and point missing | MAE: AQI: 17.23 { $\pm$ 1.12} -> 14.73 { $\pm$ 0.15}; METR-LA block: 2.96 { $\pm$ 0.18} -> 2.03 { $\pm$ 0.00}; METR-LA point: | Helped |
| Use the sensor graph rather than a fully connected graph (ablation) | GRIN · METR-LA block missing | MAE: 2.63 { $\pm$ 0.01} -> 2.03 { $\pm$ 0.00} | Helped |
| Add graph edges rather than use a graph with no edges (ablation) | GRIN · METR-LA block missing | MAE: 3.42 { $\pm$ 0.04} -> 2.03 { $\pm$ 0.00} | Helped |
Closest papers by meaning
- Motif-aware Graph Neural Networks for Networked Time Series Imputation (AAAI 2025 · similarity 0.92)
- Graph Structure Learning for Spatial-Temporal Imputation: Adapting to Node and Feature Scales (AAAI 2025 · similarity 0.90)
- Fully-Connected Spatial-Temporal Graph for Multivariate Time-Series Data (AAAI 2024 · similarity 0.90)
- Learning from Highly Sparse Spatio-temporal Data (NeurIPS 2024 · similarity 0.89)
- Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling (ICML 2024 · similarity 0.89)
- Glocal Information Bottleneck for Time Series Imputation (NeurIPS 2025 · similarity 0.88)
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.