Reproduce “Progress in Self-Certified Neural Networks” 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
Maria Perez-Ortiz, Omar Rivasplata, Emilio Parrado-Hernandez, Benjamin Guedj, John Shawe-Taylor
The authors test self-certified learning on four classification datasets. They train a probabilistic neural network posterior using all available data, with disjoint prior and certification subsets, and compute PAC-Bayes risk certificates. With at least 75% removed, certificates are much lower than held-out test-set bounds on Bioresponse and Spambase. This compares bounds, not measured test error, and is limited to these datasets and protocols.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.