Reproduce “Adaptive Mean Estimation by In-Context Learning: A Gradient-Flow Analysis” 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
Martin Eppert, Krishna Balasubramanian, Subhro Ghosh, Jason Klusowski, Yan Shuo Tan
The authors study a scalar location estimator that adapts to an unlabeled sample’s noise family. It combines mean-based and extreme-based estimates using attention features and learned gates. A staged-training theorem gives Gaussian mean-squared error of (1+o(1))/n and uniform error O(n^(-2+ε)), near the respective minimax rates. The guarantee is limited to a two-expert model under assumptions; joint-training experiments show related specialization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.