Research map / Time series / Foundation models for time series
Beyond Extrapolation: Knowledge Utilization Paradigm with Bidirectional Inspiration for Time Series Forecasting
Liu Chong, Yingjie Zhou, Hao Li, Pengyang Wang, Qingsong Wen, Ce Zhu
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 alpha residual weight (ablation) | KUP-BI on xPatch · ILI | MSE: 1.929 -> 1.366 | Improved |
| add alpha residual weight (ablation) | KUP-BI on xPatch · ETTm1 | MSE: 0.412 -> 0.352 | Improved |
| add alpha residual weight (ablation) | KUP-BI on xPatch · ETTh1 | MSE: 0.457 -> 0.431 | Improved |
| add gated fusion instead of concatenation (ablation) | KUP-BI on xPatch · ILI | MSE: 1.713 -> 1.366 | Improved |
| add gated fusion instead of concatenation (ablation) | KUP-BI on xPatch · ETTh1 | MSE: 0.411 vs 0.431 | Hurt |
| use correlation-based retrieval instead of random (ablation) | KUP-BI on xPatch · ETTh1 | MSE: 0.443 -> 0.431 | Improved |
| use ratio-style continuation instead of direct continuation (ablation) | KUP-BI on xPatch · ETTh1 | MSE: 0.462 -> 0.431 | Improved |
| use post-target continuation instead of target segment (ablation) | KUP-BI on xPatch · Exchange | MSE: 0.355 vs 0.363 | Hurt |
| use post-target continuation instead of target segment (ablation) | KUP-BI on xPatch · ETTh1 | MSE: 0.466 -> 0.431 | Improved |
| add temperature scaling tau (ablation) | KUP-BI on xPatch · ILI | MSE: 1.382 -> 1.366 | Improved |
| aggregate top-k candidates rather than k=1 (ablation) | KUP-BI on xPatch · ILI | MSE: 1.389 -> 1.366 | Improved |
| retrieval instead of prediction-based continuation (ablation) | KUP-BI on DLinear · ETTh2 | MSE: 0.551 -> 0.453 | Improved |
Closest papers by meaning
- Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting (NeurIPS 2026 · similarity 0.86)
- Sequence Complementor: Complementing Transformers for Time Series Forecasting with Learnable Sequences (AAAI 2025 · similarity 0.86)
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning (NeurIPS 2025 · similarity 0.85)
- A Closer Look at Transformers for Time Series Forecasting: Understanding Why They Work and Where They Struggle (ICML 2025 · similarity 0.85)
- ProbTS: Benchmarking Point and Distributional Forecasting across Diverse Prediction Horizons (NeurIPS 2024 · similarity 0.85)
- Unlocking the Power of Patch: Patch-Based MLP for Long-Term Time Series Forecasting (AAAI 2025 · 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.
