Reproduce “A Spectral Renormalization-Group View of 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
Charles H. Martin
The paper proposes a layerwise spectral correction that nudges network layers toward a power-law exponent of about 2. It interprets this value as balancing spectral energy across logarithmic scales and tests WW-PGD with AdamW on a three-layer MNIST MLP. Layer exponents move toward 2, while accuracy stays close to or slightly below baseline, so improved accuracy is not demonstrated.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.