What Is Unsupervised Learning? Finding Structure Without Labels
A training approach that finds patterns and structure in data with no labeled correct answers at all — letting the data's own structure guide what gets learned.
Finding patterns with no correct answers provided
Unsupervised learning finds patterns, groupings, or structure in data with no labeled correct answers provided at all — unlike supervised learning, there's no "correct output" for the model to be shown; it works purely from patterns inherent in the raw, unlabeled data itself.
What this actually looks like in practice
Common unsupervised tasks include clustering (grouping similar data points together without being told in advance what the groups should be) and dimensionality reduction (finding a simpler, more compact representation of complex data) — tasks defined by finding structure, not by predicting a specific known target value.
How this relates to self-supervised learning
Self-supervised learning — the approach behind language model pretraining — is sometimes described as a specific, particularly effective variant of unsupervised learning, since it also uses unlabeled data, but generates its own explicit training targets directly from that data's structure rather than only searching for looser, unguided patterns.
Frequently Asked Questions
What is unsupervised learning?
A training approach that finds patterns, groupings, or structure in data with no labeled correct answers provided, working purely from patterns inherent in the raw, unlabeled data itself.
How is unsupervised learning different from self-supervised learning?
Self-supervised learning also uses unlabeled data but generates its own explicit training targets directly from the data's structure, while classic unsupervised learning searches for looser patterns like clusters without a specific target to predict.