Unsupervised Learning

Unsupervised learning studies how to discover patterns and useful representations from data when the desired answers are not labeled. Common approaches include clustering and dimensionality reduction, latent-variable and generative models, and self-supervised methods that learn from relationships or predictions within the data. These methods can make large unlabeled collections useful for tasks such as image and speech understanding, scientific discovery, and anomaly detection. A central challenge is determining whether discovered structure is meaningful and stable: results depend on objectives and assumptions, and unlabeled training can still encode human choices or sensitive information.