DistDF：時序預測需聯合Wasserstein對齊

#optimal-transportdistdfdistdficlrwassersteinxiaohongshupeking-university

💡ICLR paper fixes MSE bias in TS forecasting via Wasserstein—open-source code ready to test!

⚡ 30-Second TL;DR

有什麼變化

MSE損失因標籤序列自相關性而有偏差

為什麼重要

將最優傳輸引入預測，提升對真實時序依賴的處理能力。在金融或氣象等自相關領域可提升準確度。

下一步行動

Clone the DistDF GitHub repo and replace MSE with its Wasserstein loss in your PyTorch time-series model.

誰應關注：Researchers & Academics

背景與延伸：來自公開資料，非原文內容。引用 7 個來源。

•DistDF arXiv preprint was first submitted on October 28, 2025, by lead author Hao Wang and collaborators from institutions including Peking University.[1]
•The paper demonstrates state-of-the-art forecasting performance on datasets like Weather, with average MSE improvements such as 0.255 vs. 0.277 for baseline models.[2]
•DistDF is compatible with diverse forecast models, enhancing their performance through integration with the joint Wasserstein discrepancy during gradient-based training.[3]

•DistDF formalizes training via Algorithm 1, processing historical sequences X and label sequences Y ∈ ℝ^B×T, minimizing joint-distribution Wasserstein discrepancy W_p(P_{X,Y}, P_{X,Ŷ}).[3]
•The joint Wasserstein discrepancy upper-bounds the conditional discrepancy (Lemma 3.3), enabling tractable estimation from finite samples and differentiability for optimization.[2][3]
•Inspired by domain adaptation (e.g., Courty et al., 2017), but focuses on conditional alignment in forecasting rather than marginal input distributions.[3]

DistDF will be integrated into production TS forecasting systems by mid-2026

ICLR 2026 acceptance and open-sourced code enable rapid adoption by industry users like Xiaohongshu authors.

Joint Wasserstein will become standard for autocorrelation-heavy TS tasks

Proven MSE improvements on benchmarks like Weather validate its superiority over likelihood-based methods.

2025-10

arXiv preprint v1 submitted by Hao Wang et al.

2026-02

ICLR 2026 paper acceptance announced.

Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.

AI 週報

閱讀本週精選 AI 大事摘要 →

同主題

探索 #optimal-transport

同產品

更多關於 distdf

同來源