Study proposes method to certify tail-risk claims when selective labels hide worst errors
A new paper argues that historical decisions determine which outcomes are observed, potentially concealing a predictor's largest errors, and introduces a framework for certifying tail-risk claims under such selective labels.
Read the full story at arXiv stat.ML ↗
Timeline · 1 report
- 2026-11-10 05:00 · arXiv stat.ML
Hidden Tails: Certifying Tail-Risk Claims under Selective Labels