Reproduce “PAC-Bayes unleashed: generalisation bounds with unbounded losses” 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
Maxime Haddouche, Benjamin Guedj, Omar Rivasplata, John Shawe-Taylor
HYPE derives a PAC-Bayes bound when losses lack a shared finite ceiling, provided each predictor has a data-independent finite ceiling. It uses self-bounding concentration and requires a sample-independent prior and posterior with identical support. In synthetic regression, optimizing the bound’s parameter gives a better bound than fixing it at one-half; the calculation assumes bounded inputs and outputs, and the bound may be loose.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.