Reproduce “A Geometric Modeling of Occam's Razor in Deep Learning” 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
Ke Sun, Frank Nielsen
The authors use minimum description length (MDL) to explain deep-network complexity. They count informative parameter directions locally using the Fisher information matrix’s rank, rather than all weights. Under a Gaussian prior, their derived penalty depends on rank and positive eigenvalues. Empirical local dimensionality is bounded by N(m-1) for fixed output size m. This is theoretical, not evidence of improved generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.