Bridging Contrastive Learning and Domain Adaptation: Theoretical Perspective and Practical Application

TL;DR

This paper establishes a theoretical link between contrastive learning and domain adaptation, demonstrating that minimizing contrastive losses reduces domain discrepancy and enhances class separation, with practical validation on mammography datasets.

Contribution

It provides a theoretical framework connecting contrastive learning to domain adaptation and validates it through extensive experiments on medical imaging data.

Findings

Minimizing contrastive losses decreases CMMD.

Contrastive learning improves class-separability.

Enhanced classification performance in mammography datasets.

Abstract

This work studies the relationship between Contrastive Learning and Domain Adaptation from a theoretical perspective. The two standard contrastive losses, NT-Xent loss (Self-supervised) and Supervised Contrastive loss, are related to the Class-wise Mean Maximum Discrepancy (CMMD), a dissimilarity measure widely used for Domain Adaptation. Our work shows that minimizing the contrastive losses decreases the CMMD and simultaneously improves class-separability, laying the theoretical groundwork for the use of Contrastive Learning in the context of Domain Adaptation. Due to the relevance of Domain Adaptation in medical imaging, we focused the experiments on mammography images. Extensive experiments on three mammography datasets - synthetic patches, clinical (real) patches, and clinical (real) images - show improved Domain Adaptation, class-separability, and classification performance, when minimizing the Supervised Contrastive loss.