Reproduce “CATF: A Manager-Worker Framework for Context-Aware 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
Yueqiao Chen, Dongyu Liu
CATF forecasts one target series from its history and context, routing each input to one specialist Worker. A Transformer-based Manager selects the Worker; all Workers are evaluated during training, but only the selected Worker runs at inference. Across five datasets and four horizons, CATF lowers MSE in 162 of 200 comparisons, with advantages most evident when context marks distinct regimes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.