What is the difference between supervised, self-supervised, and reinforcement learning?
Reviewed by Jason Burns, Editorial Steward · Last updated
Supervised learning trains on labeled input–output pairs, self-supervised learning creates its own labels from unlabeled data (predicting the next token, filling a masked patch), and reinforcement learning trains an agent to maximize cumulative reward through trial and error. As Richard S. Sutton & Andrew G. Barto, put it on the record: "Reinforcement learning is learning what to do—how to map situations to actions—so as to maximize a numerical reward signal."