Reproduce “CTRL: Control-Based Time Series Forecasting with LLM-Guided Residual Learning” 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
Minkyoung Kim, Daeun Ji, Yohan Lee, Beomsoo Kim, Beakcheol Jang
CTRL separates forecasting from correction: a frozen time-series backbone makes numerical forecasts, while language-model agents diagnose errors and guide a learned residual decoder. At test time, input statistics can trigger updates to trend and seasonal guidance without test labels. On ETTm2, CTRL with DLinear reports MSE 0.197 versus 0.245 for LLMTime, but results vary by dataset and horizon.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.