SINCERE: Supervised Information Noise-Contrastive Estimation REvisited

TL;DR

This paper revisits supervised contrastive learning, identifying issues with existing methods and proposing SINCERE, a theoretically justified loss that improves class separation and transfer learning performance.

Contribution

The paper introduces SINCERE, a new supervised InfoNCE loss that removes intra-class repulsion and is backed by theoretical analysis.

Findings

SINCERE improves class separation in embeddings.

SINCERE enhances transfer learning accuracy.

Theoretical bounds relate SINCERE to KL divergence.

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. This SupCon loss has been widely-used due to reports of good empirical performance. 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 space. This problematic intra-class repulsion gets worse as the number of images sharing one class label increases. We propose the Supervised InfoNCE REvisited (SINCERE) loss as a theoretically-justified supervised extension of InfoNCE that eliminates intra-class repulsion. Experiments show that SINCERE leads to better separation of embeddings from different classes and improves transfer learning classification accuracy. We additionally utilize probabilistic modeling to derive an information-theoretic bound that relates SINCERE loss to the symmeterized KL divergence between data-generating distributions for a target class and all other classes.