Contrastive Learning

The idea of contrastive learning is to learn representations where similar input will have closer representations and vice versa, without explictly modeling the similarity.