Reproduce “Kernel Renormalization in Bayesian Deep Neural Networks: the Equivalent Wishart Ansatz in the Proportional Regime” 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
Paolo Baglioni, Christian Keup, Vincenzo Zimbardo, Rosalba Pacelli, Alessandro Vezzani, Raffaella Burioni, Pietro Rotondo
The authors present the Equivalent Wishart Ansatz, an approximate theory for predicting finite-width behavior in Bayesian deep nonlinear networks. It models fluctuations in layer-wise similarity matrices with self-consistent scalar factors, extending the infinite-width kernel description. Comparisons with posterior sampling show that it tracks generalization errors on fully connected networks, especially around load α≲1; deviations arise at larger loads and depths.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.