Conformalized Exceptional Model Mining: Telling Where Your Model Performs (Not) Well

Xin Du, Sikun Yang(✉), Wouter Duivesteijn, Mykola Pechenizkiy · 2025

Read Conformalized Exceptional Model Mining: Telling Where Your Model Performs (Not) Well by Xin Du, Sikun Yang(✉), Wouter Duivesteijn, Mykola Pechenizkiy (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-2508-15569.txt.

Description

A model can score well on average yet fail badly for one group of cases, and finding those groups matters. Xin Du and co-authors build a framework for that search in a 2025 paper. Their method joins two tools: conformal prediction, which gives firm bounds on doubt, and exceptional model mining, which hunts for odd subgroups. High-stakes fields like health and finance need this kind of check.

A new model class scores how unsure a model is inside each subgroup. A fresh quality measure then flags groups where doubt runs high or low in odd ways. Tests on varied data sets showed the method finds clear, readable subgroups. The authors say such maps of weak spots can make AI use more honest and safe.

Catalog record

CREATORXin Du, Sikun Yang(✉), Wouter Duivesteijn, Mykola Pechenizkiy
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-15569

Browse the full library · About the Tower of Babel