Selective Omniprediction and Fair Abstention

TL;DR

We propose new learning algorithms for building selective classifiers, which are predictors that are allowed to abstain on some fraction of the domain…

Opening excerpt from the authors’ abstract. source