Reproduce “PAC-Bayes Analysis Beyond the Usual Bounds” 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
Omar Rivasplata, Ilja Kuzborskij, Csaba Szepesvari, John Shawe-Taylor
The paper formulates PAC-Bayes bounds allowing learner and reference distributions to depend on training samples. A kernel maps samples to hypothesis distributions. The theorem bounds a convex function of population and empirical risks using KL divergence and an exponential-moment term that must be controlled. For linear regression it gives an almost-sure bound for unbounded square loss under a covariance-eigenvalue condition.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.