Reproduce “Conformal Selective Prediction with General Risk Control” 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
Tian Bai, Ying Jin
SCoRE gives models a way to abstain from predictions when their expected bounded risk is too high. It uses labeled calibration examples to construct risk-adjusted e-values, then applies testing rules to control marginal deployment risk or average risk among selected cases. In simulations, it tracked nominal targets and outperformed concentration baselines in selection power; guarantees require bounded risks and exchangeability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.