SINCERE: Supervised Information Noise-Contrastive Estimation REvisited

arXiv:2309.14277v2 Announce Type: replace-cross
Abstract: The information noise-contrastive estimation (InfoNCE) loss function provides the basis of many self-supervised deep learning methods due to its strong empirical results and theoretic motivation. Previous work suggests a supervised contrastive (SupCon) loss to extend InfoNCE to learn from available class labels. However, in this work we find that the prior SupCon loss formulation has questionable justification because it can encourage some images from the same class to repel one another in the learned embedding …

abstract arxiv class cs.cv cs.lg deep learning function information labels learn loss motivation noise the information type work