Contrastive learning methods are a class of self-supervised learning techniques aimed at learning a representation where similar data points are close to each other in the feature space and dissimilar ones are farther apart.
This is usually achieved through a contrastive loss that uses pairs of similar (positive) and dissimilar (negative) examples.
Well-known approaches include SimCLR, MoCo, or BYOL, which use different strategies to generate positive and negative examples and optimize the representations.
Contrastive learning enables the creation of robust feature vectors without explicit labels, which can be used for many downstream tasks such as classification or clustering.
The quality of the representation depends on the selection of positive and negative pairs as well as the architecture of the underlying model.