Research map / Time series / Causal and generative time series
Quadratic Direct Forecast for Training Multi-Step Time-Series Forecast Models
Eric Wang, Licheng Pan, Yuan Lu, Zi Ciu Chan, Tianqiao Liu, Shuting He, Zhixuan Chu, Qingsong Wen, Haoxuan Li, Zhouchen Lin
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 heterogeneous (diagonal) task weights to DF (ablation) | TQNet · ETTh1 (avg) | MSE: 0.449 -> 0.443 | Improved |
| add label-autocorrelation (off-diagonal) weights to DF (ablation) | TQNet · ETTh1 (avg) | MSE: 0.449 -> 0.442 | Improved |
| combine both diagonal and off-diagonal weighting (ablation) | TQNet · ETTh1 (avg) | MSE: 0.442 -> 0.431 | Improved |
| use QDF instead of DF objective | TQNet · ECL (avg) | MSE: 0.175 -> 0.165 | Improved |
| use QDF instead of DF objective | TQNet · PEMS08 (avg) | MSE: 0.139 -> 0.120 | Improved |
| use QDF instead of FreDF/Time-o1 objective | TQNet · ETTh1 (avg) | MSE: 0.432 -> 0.431 | Improved |
| use QDF instead of Koopman objective | TQNet · ETTm1 (avg) | MSE: 0.595 -> 0.371 | Improved |
| use QDF instead of Soft-DTW objective | TQNet · ECL (avg) | MSE: 0.623 -> 0.165 | Improved |
| use QDF bilevel update instead of meta-learning algorithms for weighting matrix (ablation) | TQNet · ECL (T=720) | MSE: 0.204 -> 0.202 | Improved |
| increase inner-loop updates N_in from 0 to 1 (ablation) | QDF · hyperparameter sweep | accuracy | Improved |
| split data into K=3 instead of more subsets (ablation) | QDF · hyperparameter sweep | accuracy | Mixed |
| add QDF to a PDF backbone on ECL | PDF · ECL (avg) | MSE: 0.198 -> 0.194 | Improved |
Closest papers by meaning
- FreDF: Learning to Forecast in the Frequency Domain (ICLR 2025 · similarity 0.86)
- Time-o1: Time-Series Forecasting Needs Transformed Label Alignment (NeurIPS 2025 · similarity 0.83)
- Analysing Multi-Task Regression via Random Matrix Theory with Application to Time Series Forecasting (NeurIPS 2024 · similarity 0.83)
- Structured Matrix Basis for Multivariate Time Series Forecasting with Interpretable Dynamics (NeurIPS 2024 · similarity 0.82)
- Multivariate Time Series Forecasting needs Cross Variable Loss (NeurIPS 2026 · similarity 0.82)
- SEMPO: Lightweight Foundation Models for Time Series Forecasting (NeurIPS 2025 · similarity 0.81)
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.
