SLOVA: Uncertainty estimation using single label one-vs-all classifier

Bartosz Wójcik, Jacek Grela, Marek Śmieja, Krzysztof Misztal, Jacek Tabor

Abstract: Highlights•We present a novel method for classification uncertainty assessment.•We test our method on in-Distribution, shifted and Out-Of-Distribution data.•The results are comparable to state of the art methods.•The method is post-hoc and requires little modification of a standard network.

External IDs:doi:10.1016/j.asoc.2022.109219