Reproduce “Multivariate Time Series Forecasting needs Cross Variable Loss” 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
Kuiye Ding, Yifan Hu, Hao Xue, Hanchen Wang
CvLoss adds a graph-based penalty to multivariate forecasting to coordinate residuals across variables and time patches. It compares predicted differences between connected variable-patches with their true differences, alongside the usual point-wise loss. Across 114 fixed-backbone dataset-and-metric comparisons, it improves 111, ties 2, and loses 1; mean reductions are 3.39% for MSE and 3.61% for MAE, not a universal gain.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.