Study finds growing label lists silently alter conformal decisions in vision-language classifiers

A new arXiv paper shows that as vision-language classifiers learn new labels from text, the growth of their label registries silently changes conformal prediction decisions, and proposes auditing and preservation methods.

Read the full story at arXiv cs.CV ↗

Timeline · 1 report

1. 2026-11-06 05:00 · arXiv cs.CV
   Coverage Hides Churn: Auditing and Preserving Conformal Decisions as VLM Label Registries Grow