Reproduce “Infinitely wide limits for deep Stable neural networks: sub-linear, linear and super-linear activation functions” 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
Alberto Bordino, Stefano Favaro, Sandra Fortini
The authors derive limits for deep networks with heavy-tailed Stable weights. They track how activation growth changes transformed weight tails, then use a generalized central limit theorem to determine normalization and limiting law. Under sequential width growth, ReLU requires a logarithmic correction, while a shallow cubic activation changes the stability index to α/3. Joint growth and general multiple-input limits remain unestablished.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.