Self-Supervised Learning

Self-supervised learning derives supervisory signals from unlabeled data. A pretext task creates the learning objective, while contrastive learning separates representations of unrelated examples.

Japanese terms

1. Self-supervised learning — 自己教師じこきょうしあり学習がくしゅう: Learning in which supervisory signals are derived from the structure of the input data itself.
2. Pretext task — 代理だいりタスク: A task constructed from unlabeled data to provide a learning signal for representation learning.
3. Contrastive learning — 対照学習たいしょうがくしゅう: Learning representations by bringing related examples closer and separating unrelated examples.

Related topics

Self-supervised learning is central to the pretraining of many foundation models.