AI Glossary Term

What does “Self-Supervised Learning” mean?

Definition

Learning representations from unlabeled data by predicting masked or transformed parts.

Browse the full AI Glossary Explore all free AI guides

How Self-Supervised Learning works

Example

A language model is shown sentences with the next word hidden and learns by guessing it. The text itself supplies the answers, so nobody has to label billions of examples.

Why it matters

Self-supervision makes use of vast unlabeled data, which is the basis of modern language and many vision models.

Common questions

Self-supervised vs unsupervised?

Both avoid human labels. Self-supervised creates its own prediction task from the data, such as filling in missing parts.

Which models use it?

Large language models and many image and audio models are pre-trained this way.

Does it still need labels?

Not for pretraining, though later stages may use labeled examples.

Unsupervised LearningLearning patterns from unlabeled data without explicit target outputs.Data LineageA record of where data came from, how it was transformed, and where it is used.Self-PlayA training setup where a model improves by generating data through interactions or competitions with copies of itself.Continual LearningTraining approaches that let a model keep learning from new data without forgetting prior knowledge.Data LabelingThe process of assigning tags or target outputs to raw data for supervised learning.Intent ClassificationPredicting the user's purpose from a text query to route it correctly.

Learn more in our free guides

Deep Learning Reinforcement Learning Machine Learning Basics Supervised Learning

Test yourself

Deep Learning Quiz Reinforcement Learning Quiz Machine Learning Basics Quiz