Reproduce “Born Discrete, Made Smooth: Variational Formulation of Shallow Neural Networks” 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
Matej Benko, Pierre Bousquet, Iwona Chlebicka, Błażej Miasojedow
The authors represent shallow-network training with signed densities over hidden-unit parameters. Weighted squared-density and parameter-space smoothness penalties make their continuum objective globally convex. Under stated assumptions, its unique minimizer exists and is locally C²˒ˢ for every s<1. On Diabetes, held-out RMSE is 52.00 versus 63.28 for a neural-network baseline, not a generalization guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.