Unsupervised Learning

A type of machine learning that looks for previously undetected patterns in a data set with no pre-existing labels.

How it works

Unsupervised learning algorithms discover structure in data without any labelled outputs. Clustering algorithms (K-Means, DBSCAN, hierarchical clustering) group similar examples together. Dimensionality reduction techniques (PCA, t-SNE, UMAP) compress data into lower-dimensional representations that preserve structure. Generative models (VAEs, GANs, diffusion models) learn the underlying distribution of the data to generate new samples. Anomaly detection identifies examples that don't fit the learned distribution.

Why it matters

Unsupervised learning unlocks value from the vast majority of real-world data, which comes without labels. It is the basis of the pre-training that powers all modern LLMs — next-token prediction is an unsupervised technique that requires no human annotation. In practice, unsupervised learning discovers market segments, detects novel fraud patterns, compresses high-dimensional data for visualisation, and generates synthetic training data that supplements limited labelled datasets.