Conformal Prediction Sets with Improved Conditional Coverage using Trust Scores

Jivat Neet Kaur ∗ Michael I. Jordan ∗† Ahmed Alaa ∗ · 2025

Read Conformal Prediction Sets with Improved Conditional Coverage using Trust Scores by Jivat Neet Kaur ∗ Michael I. Jordan ∗† Ahmed Alaa ∗ (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-10139.txt.

Description

Jivat Neet Kaur and co-authors improved prediction sets with trust scores in a 2025 paper. Conformal prediction wraps a model's guess in a set that should hold the truth at a set rate overall. The rub is that the rate can sag on the very cases where the model is wrong yet sounds sure.

Their method adds a trust score that flags such overconfident spots. Coverage is then aimed where it matters most, at points the classifier gets wrong with high confidence. Exact per-point coverage is known to be out of reach, so the team target the worst gap. The result is sets that fail more fairly.

Catalog record

CREATORJivat Neet Kaur ∗ Michael I. Jordan ∗† Ahmed Alaa ∗
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE58 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-10139

Browse the full library · About the Tower of Babel