Contrastive representation learning is an approach in the field of self-supervised learning that aims to create a meaningful embedding (representation) of data.
Pairs of similar (positive) and different (negative) examples are formed to train a model that brings similar examples closer together in the feature space and pushes dissimilar ones further apart.
Typical methods use a contrastive function such as the triplet loss or the InfoNCE loss to control the distance between the representations.
This technique is often used in image processing, speech processing, and other areas to learn robust features without explicit labels.