Reproduce “Two hidden layer suffice: the caterpillar net” 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
Jakob Paul Zimmermann
The Caterpillar Net constructs an exact two-hidden-layer ReLU network for the maximum of N real inputs for every N≥3, on all of R^N. It uses caterpillar-shaped trees and signed geometric cancellations instead of recursively nesting pairwise maxima. Its second-layer width is 2^{O(N log N)}, versus 2^{Θ(N²)} for a recursive baseline; extraction remains intractable for large N.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.