density-ratio estimation via classification is a cornerstone of unsupervised
learning. It has provided the foundation for state-of-the-art methods in
representation learning and generative modelling, with the number of use-cases
continuing to proliferate. However, it suffers from a cri