Self-Supervised Learning

Self-supervised learning creates training signals from the input itself — predict masked tokens, next tokens, or augmented views — enabling pretraining at web scale without manual labels.

Modern foundation models are primarily self-supervised learners.

← Back to AI Glossary