Conditional validity of inductive conformal predictors

About

Conformal predictors are set predictors that are automatically valid in the sense of having coverage probability equal to or exceeding a given confidence level. Inductive conformal predictors are a computationally efficient version of conformal predictors satisfying the same property of validity. However, inductive conformal predictors have been only known to control unconditional coverage probability. This paper explores various versions of conditional validity and various ways to achieve them using inductive conformal predictors and their modifications.

Vladimir Vovk · 2012

Related benchmarks

Task	Dataset	Result
Conformal Prediction	ImageNet	Average Prediction Set Size4.2	63
Conformal Prediction	CIFAR-100	Avg Prediction Set Size47.2	32
Classification	ImageNet	WUC0.019	24
Classification	ImageNet V2	WUC0.042	24
Classification	CIFAR-100	WUC0.024	24
Image Classification	ImageNet	Coverage65.8	24
Image Classification	ImageNet V2	Coverage (Cov)70.8	24
Image Classification	CIFAR-100	Coverage58.2	24
Conformal Prediction	CIFAR-100 (five repeated splits)	Class Coverage58.2	24
Text Classification	WOS-46985	Coverage62.1	24