One-Sentence Definition
Unsupervised learning is a branch of machine learning where a model finds patterns, groupings, or structure in data without being given labeled examples or explicit instructions about what to look for.
How It Works
In supervised learning, every training example comes with a label: this email is spam, this image is a cat. Unsupervised learning removes the labels entirely. The model receives raw data and must discover structure on its own.
The most common technique is clustering. Algorithms like k-means, DBSCAN, and Gaussian mixture models group data points that are similar to each other. A retailer might feed purchase histories into a clustering algorithm and discover that its customers naturally fall into five distinct segments -- without ever defining those segments in advance.
Dimensionality reduction is another core technique. Methods like principal component analysis (PCA) and t-SNE compress high-dimensional data into fewer dimensions while preserving meaningful relationships. This is essential for visualization and for making downstream models more efficient. When researchers plot millions of data points on a 2D chart and see clear clusters, they are usually using dimensionality reduction.
Anomaly detection is a third major application. By learning what normal data looks like, unsupervised models can flag outliers -- a fraudulent credit card transaction, a malfunctioning sensor on a factory floor, or unusual network traffic that signals a cyberattack.
More recently, self-supervised learning -- the technique behind large language models like GPT-4 and Claude -- blurs the line. These models train on unlabeled text by predicting masked or next tokens, which is technically unsupervised, but the training objective creates implicit labels from the data itself. Some researchers classify self-supervised learning as a subset of unsupervised learning; others treat it as its own category.
Why It Matters
Most of the world's data is unlabeled. Labeling is expensive, slow, and sometimes impossible -- you cannot label customer segments that you have not yet discovered. Unsupervised learning lets organizations extract value from data that would otherwise sit unused.
In practice, unsupervised learning drives customer segmentation at companies like Spotify and Netflix, powers anomaly detection at financial institutions like JPMorgan Chase, and enables the pretraining phase of nearly every large language model. Google uses unsupervised clustering to organize search results, and cybersecurity firms like CrowdStrike use it to detect novel threats that have no prior labeled examples.
Key Takeaway
Unsupervised learning discovers hidden structure in data without labels, making it essential for clustering, anomaly detection, and the pretraining stage of modern AI systems.
Part of the AI Weekly Glossary.