Abstract

We study optimal conformity measures for various criteria of efficiency of set-valued classification in an idealised setting. This leads to an important class of criteria of efficiency that we call probabilistic and argue for; it turns out that the most standard criteria of efficiency used in literature on conformal prediction are not probabilistic unless the problem of classification is binary. We consider both unconditional and label-conditional conformal prediction.
Original languageEnglish
Pages (from-to)21-46
Number of pages26
JournalAnnals of Mathematics and Artificial Intelligence
Volume81
Early online date14 Mar 2017
DOIs
Publication statusPublished - Oct 2017

Cite this