Reproduce “Online Conformal Prediction Beyond Feedback” 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
Joar Skalse, Edoardo Pona, Osvaldo Simeone, Nicola Paoletti
OCPQ lets a classifier either issue prediction sets or query labels, but not both. It uses queried labels to estimate how candidate thresholds would have performed and update threshold weights. For a fixed black-box classifier on an oblivious, possibly non-i.i.d. stream, it guarantees expected coverage of at least β−O(T−1/3), with O(T2/3) expected regret and a T−1/3 query fraction; queries count as uncovered.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.