Research map / Time series / Long-term forecasting
Discrete Graph Structure Learning for Forecasting Multiple Time Series
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 |
|---|---|---|---|
| Learn graph probabilities jointly with the forecaster using a parameterized unilevel objective rather than LDS bilevel optimization | GTS forecasting model · METR-LA | MAE: 2.75 -> 2.64 | Helped |
| Learn graph probabilities jointly with the forecaster using a parameterized unilevel objective rather than LDS bilevel optimization | GTS forecasting model · PEMS-BAY | MAE: 1.33 -> 1.32 | Helped |
| Learn graph probabilities jointly with the forecaster using a parameterized unilevel objective rather than LDS bilevel optimization | GTS forecasting model · PMU | MAE (×10⁻³): 0.49 -> 0.26 | Helped |
| Replace the DCRNN forecaster with T-GCN while retaining GTS graph learning (ablation) | GTS · METR-LA | MAE: 2.64 -> 2.74 | Hurt |
| Replace the DCRNN forecaster with T-GCN while retaining GTS graph learning (ablation) | GTS · PEMS-BAY | MAE: 1.32 -> 1.35 | Hurt |
| Replace the DCRNN forecaster with T-GCN while retaining GTS graph learning (ablation) | GTS · PMU | MAE (×10⁻³): 0.26 -> 0.35 | Hurt |
| Increase graph-prior regularization or use a smaller kNN neighborhood (ablation) | GTS · PMU, 15-minute forecasting | MAE: all around 2.4e-4 | No clear effect |
| Use graph-prior regularization to move the learned graph toward the supplied prior (ablation) | GTS · METR-LA and PEMS-BAY | forecasting quality | Hurt |
| Use the supplied graph directly instead of learning a forecast-oriented structure (ablation) | GTS · METR-LA and PEMS-BAY | forecasting quality | Hurt |
| Use unilevel graph learning instead of LDS bilevel optimization | GTS · METR-LA training | training time: approximately three times | Helped |
Closest papers by meaning
- Graph-based Time Series Clustering for End-to-End Hierarchical Forecasting (ICML 2024 · similarity 0.89)
- On the Interplay between Graph Structure and Learning Algorithms in Graph Neural Networks (ICML 2025 · similarity 0.87)
- Multi-Scale Dynamic Graph Learning for Time Series Anomaly Detection (Student Abstract) (AAAI 2024 · similarity 0.87)
- HyperIMTS: Hypergraph Neural Network for Irregular Multivariate Time Series Forecasting (ICML 2025 · similarity 0.86)
- Structured Matrix Basis for Multivariate Time Series Forecasting with Interpretable Dynamics (NeurIPS 2024 · similarity 0.86)
- Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling (ICML 2024 · 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.