Reproduce “Exact Robust Conformal Calibration With Suspect Labels and Corruption Budgets” 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
Daniel Huang
The method calibrates prediction sets with possibly incorrect labels by using a suspect set and an error budget. It selects the smaller of a global rank-shift threshold and a pessimistic partial-label threshold, using two order-statistic selections after score evaluation. In the CIFAR-10N audit, mean set size was 1.547, versus 1.311 for a clean audit; results apply in this setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.