Reproduce “A Widely Applicable Bayesian Information Criterion” 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
Sumio Watanabe
WBIC estimates Bayesian evidence for singular models by averaging training loss under a posterior softened at inverse temperature 1/log n, without requiring the true distribution or real log canonical threshold. Under stated conditions, its leading asymptotic terms match free energy. In reduced-rank regression, it selected the true rank in all 100 simulated datasets, a result specific to that experiment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.