SINCERE: Supervised Information Noise-Contrastive Estimation REvisited
This addresses a theoretical flaw in a widely-used loss function for supervised contrastive learning, with incremental improvements for practitioners in deep learning.
The paper tackled the problem of intra-class repulsion in the supervised contrastive loss (SupCon), which can cause images from the same class to repel each other in embeddings, and proposed SINCERE loss as a theoretically-justified alternative that improves separation of embeddings and transfer learning accuracy.
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.