Reproduce “Conditionally Whitened Generative Models for Probabilistic Time Series Forecasting” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yanfeng Yang, Siwei Chen, Pingping Hu, Zhaotong Shen, Yingjie Zhang, Zhuoran Sun, Shuai Li, Ziqi Chen and 1 more
CW-Gen improves probabilistic time-series forecasting by conditioning generation on an estimated future mean and multivariate covariance. JMCE estimates these quantities from history, and conditional whitening recenters and decorrelates future residuals before a diffusion or flow model learns them. On ETTh1, Diffusion-TS mean CRPS falls from 0.626 to 0.445; across tested ETTh1 comparisons, CW wins 22 of 24, not universally.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.