Contrastive Learning
Learning by pulling related items together and pushing others apart.
Contrastive learning trains representations so that matching pairs (e.g., an image and its caption) land close together while mismatched pairs are pushed apart. It is the objective behind CLIP and many self-supervised vision models.