Research map / Time series / Causal and generative time series
DistDF: Time-series Forecasting Needs Joint-distribution Wasserstein Alignment
Eric Wang, Licheng Pan, Yuan Lu, Zhixuan Chu, Xiaoxi Li, Shuting He, Zi Ciu Chan, 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 |
|---|---|---|---|
| train with DistDF instead of MSE (DF) | Fredformer · ECL | MSE: 0.191 -> 0.173 | Improved |
| train with DistDF instead of MSE (DF) | TimeBridge · ETTh1 | MSE: 0.442 -> 0.434 | Improved |
| train with DistDF instead of Time-o1 | TimeBridge · ETTm1 | MSE: 0.383 -> 0.383 | No clear effect |
| use shape-alignment objectives (Dilate, Soft-DTW) instead of MSE | TimeBridge · ECL, ETTh1 | MSE: 0.176 -> 0.173 (Soft-DTW ECL); 0.442 -> 0.452 (Soft-DTW ETTh1) | Mixed |
| use likelihood-based decorrelation (FreDF, Time-o1) instead of MSE | TimeBridge · ECL | MSE: 0.176 -> 0.175 | Improved |
| add mean alignment to DF (ablation) | DF · ETTh1 | MSE: 0.447 -> 0.435 | Improved |
| add covariance alignment to DF (ablation) | DF · ETTh1 | MSE: 0.447 -> 0.439 | Improved |
| combine mean and covariance alignment (ablation) | DistDF · ETTh1 | MSE: 0.435 -> 0.430 | Improved |
| replace Bures-Wasserstein with EMD, MMD or KL (ablation) | DistDF · ETTh1, TimeBridge | MSE: 0.433 -> 0.441 (EMD) | Hurt |
| any joint-distribution discrepancy instead of MSE only (ablation) | DistDF variants · ETTm1, ETTh1, ECL, Weather | MSE | Improved |
| raise discrepancy weight gamma from 0 to 0.005 (ablation) | DistDF, TimeBridge · ECL | MSE: 0.176 -> 0.172 | Improved |
| use discrepancy only (gamma=1, no MSE term) (ablation) | DistDF, Fredformer · ECL | MSE: 0.191 -> 0.239 | Hurt |
Later papers that compared with this method
1 later accepted paper reports DistDF as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Paloma: Phase-Conditioned Residual Modulation for Time Series Forecasting (NeurIPS 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Time-o1 — Time-o1: Time-Series Forecasting Needs Transformed Label Alignment (NeurIPS 2025) · better on 4 of 4 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
- 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 0.84)
- Conditional Diffusion Model with Nonlinear Data Transformation for Time Series Forecasting (ICML 2025 · similarity 0.84)
- S4M: S4 for multivariate time series forecasting with Missing values (ICLR 2025 · similarity 0.83)
- Learning Pattern-Specific Experts for Time Series Forecasting Under Patch-level Distribution Shift (NeurIPS 2025 · similarity 0.83)
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning (NeurIPS 2025 · similarity 0.83)
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.
