Reproduce “Invertible Graph Aggregation Can Increase ReLU Readout Width” 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
Hiroyasu Usami, Masanori Tsukamoto, Kohei Hashiguchi
The authors show one graph-aggregation step can make labels harder for a shared rowwise ReLU classifier, although every feature row remains distinct. On a constructed tree task, minimum zero-error width rises from 1 on raw features to N after averaging, for every even N ≥ 4. Benchmark outcomes are fitted-procedure comparisons, not a general capacity claim.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.