Reproduce “Evaluating Epistemic Uncertainty: Beyond OOD Detection and Active 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
Jakub Paplhám, Willem Waegeman, Eyke Hüllermeier, Vojtěch Franc
In a closed-form regression sandbox, the authors show that five Bayes-optimal uncertainty rules reject different inputs at 80% coverage. They define epistemic uncertainty as regret relative to an oracle and unify risk- and regret-limited selection. On separate CIFAR-10 OOD and CIFAR-10H regret evaluations, DDU and Evidential rankings invert; this is not a same-dataset comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.