In this paper we develop a principled, probabilistic, unified approach to non-standard classification tasks, such as semi-supervised, positive-unlabelled, multi-positive-unlabelled and noisy-label learning. We train a classifier on the given labels to predict the label-distribution. We then infer the underlying class-distributions by variationally optimizing a model of label-class transitions.
@article{arxiv.2006.09046,
title = {Probabilistic Decoupling of Labels in Classification},
author = {Jeppe Nørregaard and Lars Kai Hansen},
journal= {arXiv preprint arXiv:2006.09046},
year = {2020}
}