Criteria of efficiency for conformal prediction
This work addresses theoretical efficiency criteria for conformal prediction, which is incremental as it refines existing methods without introducing new paradigms.
The paper investigates optimal conformity measures for classification efficiency in conformal prediction, finding that standard criteria are not probabilistic except in binary cases, and examines both unconditional and label-conditional settings.
We study optimal conformity measures for various criteria of efficiency of classification in an idealised setting. This leads to an important class of criteria of efficiency that we call probabilistic; 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.