Reproduce “Sharp Generalization for Nonparametric Regression in Interpolation Space by Shallow Over-Parameterized Neural Networks with Channel Attention” 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
Yingzhen Yang, Ping Li
The paper establishes a clean population-risk guarantee for a shallow ReLU network trained with a fixed covariance-based channel-reweighting matrix. Built from initialized features and independent auxiliary covariates, the matrix remains fixed while both layers train. Under stated spherical-input, target, and size conditions, polynomial kernel spectra yield a minimax-optimal rate, also attained by kernel gradient descent on the same target ball.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.