Research map / Time series / Long-term forecasting
Sparse Graph Learning from Spatiotemporal Time Series
Andrea Cini, Daniele Zambon, Cesare Alippi
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 |
|---|---|---|---|
| Add the Fréchet-mean graph baseline for score gradients (ablation) | BES and SNS graph learning on GPVAR · GPVAR graph identification | validation MAE convergence | Improved |
| Use SNS's fixed-sparsity prior instead of BES (ablation) | Graph learning on GPVAR · GPVAR graph identification | validation MAE convergence | Improved |
| Add the local-cost surrogate objective (ablation) | BES and SNS graph learning · GPVAR graph identification | validation MAE convergence | Improved |
| Add dummy nodes to allow adaptive node degrees in SNS (ablation) | SNS graph learning · GPVAR graph identification | validation MAE convergence | Improved |
| Increase the surrogate's local-cost weighting while decreasing the global-cost weighting (ablation) | BES graph learning · GPVAR graph identification | validation MAE convergence | Improved |
| Overestimate the maximum SNS neighborhood size (ablation) | SNS graph learning · GPVAR graph identification | convergence to optimal graph | Hurt |
| Set SNS neighborhood size to K=30 (ablation) | SNS graph learning · GPVAR graph identification | convergence to optimal graph | Hurt |
| Use score-function estimation with baseline and surrogate instead of path-wise or straight-through estimation (ablation) | BES graph learning · GPVAR joint graph and model learning, simpler setting | forecasting accuracy | Mixed |
| Use score-function estimation with baseline and surrogate instead of path-wise or straight-through estimation (ablation) | BES graph learning · GPVAR joint graph and model learning, harder setting | forecasting accuracy | Improved |
| Learn a sparse graph with SNS instead of using a random adjacency | Full attention architecture · PEMS-BAY, 15-minute horizon | MAE: 1.327 -> 1.317 | Improved |
| Learn a sparse graph with SNS instead of using a random adjacency | Full attention architecture · PEMS-BAY, 30-minute horizon | MAE: 1.636 -> 1.620 | Improved |
| Learn a sparse graph with SNS instead of using a random adjacency | Full attention architecture · PEMS-BAY, 60-minute horizon | MAE: 1.897 -> 1.873 | Improved |
Closest papers by meaning
- Learning from Highly Sparse Spatio-temporal Data (NeurIPS 2024 · similarity 0.84)
- Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling (ICML 2024 · similarity 0.84)
- Towards Dynamic Spatial-Temporal Graph Learning: A Decoupled Perspective (AAAI 2024 · similarity 0.84)
- Over-squashing in Spatiotemporal Graph Neural Networks (NeurIPS 2025 · similarity 0.83)
- STRAP: Spatio-Temporal Pattern Retrieval for Out-of-Distribution Generalization (NeurIPS 2025 · similarity 0.83)
- Two Heads Are Better Than One: Boosting Graph Sparse Training via Semantic and Topological Awareness (ICML 2024 · similarity 0.82)
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.
