Reproduce “From Fixed Defaults to Calibrated Activation Policies for Long-Term 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
Shijie Xu, Boyan Tang, Kai He, Jianghua Wu
The authors develop a forecasting policy that reuses completed experiments to choose activation functions and placement for new tasks, retaining a dependable global configuration unless input characteristics support an override. Across 51 Informer dataset-horizon tasks, it reduced mean squared error by 7.86% versus fixed GELU and improved 72.55% of tasks; the conditional override’s 0.89% added gain has an interval crossing zero.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.