Research map / Time series / Long-term forecasting
Winformer: Transcending Pairwise Similarity for Time-series Generation
Haoyi Zhou, Xin Xue, Tianyu Chen, lanhao li, Lijun Sun, Jianxin Li
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 TimeDP with Winformer window-wise attention | cross-domain time-series diffusion · 12 datasets (length 168) | MMD: average 10.67% MMD decrease vs TimeDP; best on 10 of 12 datasets | Improved |
| use window-wise alignment instead of patch-wise (ablation) | Winformer · Traffic | MMD: 0.109 -> 0.071 | Improved |
| use window-wise alignment instead of patch-wise (ablation) | Winformer · Taxi | MMD: 0.195 -> 0.085 | Improved |
| use window-wise alignment instead of point-wise (ablation) | Winformer · Traffic | MMD: 0.073 -> 0.071 | Improved |
| use window-wise alignment instead of patch-wise (ablation) | Winformer · Wind | MMD: 0.033 -> 0.034 | No clear effect |
| use window-wise alignment instead of patch-wise and point-wise (ablation) | Winformer · 12 datasets, MMD and K-L | best scores count: 15 best scores vs 8 (patch) and 6 (point) | Improved |
| initialize alignment kernel with DCT basis instead of random (ablation) | Winformer · 12 datasets (MMD) | MMD: e.g. Solar 0.039 -> 0.035; Wind 0.025 -> 0.034 | Mixed |
| use Winformer on periodic datasets (solar, traffic) versus trend-dominated (wind, temperature) | Winformer · Wind, Temperature | MMD: Wind 0.025 (TimeDP) vs 0.034; Temperature 0.219 vs 0.230 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TimeDP — TimeDP: Learning to Generate Multi-Domain Time Series with Domain Prompts (AAAI 2025) · better on 7 of 10 shared results
- Diffusion-TS — Diffusion-TS: Interpretable Diffusion for General Time Series Generation (ICLR 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TimeMixer++: A General Time Series Pattern Machine for Universal Predictive Analysis (ICLR 2025 · similarity 0.87)
- TimePoint: Accelerated Time Series Alignment via Self-Supervised Keypoint and Descriptor Learning (ICML 2025 · similarity 0.86)
- Time Series Generation Under Data Scarcity: A Unified Generative Modeling Approach (NeurIPS 2025 · similarity 0.86)
- Introducing Spectral Attention for Long-Range Dependency in Time Series Forecasting (NeurIPS 2024 · similarity 0.86)
- TimeTok: Granularity-Controllable Time-Series Generation via Hierarchical Tokenization (NeurIPS 2026 · similarity 0.85)
- Peri-midFormer: Periodic Pyramid Transformer for Time Series Analysis (NeurIPS 2024 · similarity 0.85)
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.
