Unsupervised Learning

Learning patterns from unlabeled data, the model discovers hidden structure without being told the 'right answer'. Includes clustering, dimensionality reduction, and anomaly detection. A fourth classic task is association-rule mining (Apriori), the market-basket analysis behind frequently-bought-together.

Key terms

ClusteringDimensionality reductionAnomaly detectionK-MeansAssociation rules (Apriori)

Learn these first

← Machine Learning (ML)

Where you meet it in the real world

Customer segmentation, anomaly detection, topic modeling, data compression

Deep dive · 6 min

Unsupervised and Self-Supervised Learning

Labels are expensive; data is not. Unsupervised learning finds structure in unlabeled data. Clustering groups similar items: customer segments, topic groupings, anomaly detection by distance from every cluster. Dimensionality reduction compresses thousands of variables into a few meaningful ones. Embeddings map items into a space where distance means similarity, which is what powers semantic search and recommendation engines.

The trick that ate the field

Self-supervised learning manufactures labels from the data itself: hide part of the input, train the model to predict the hidden part. Hide the next word of a sentence and you get the next-token objective that trained GPT and every other large language model. Nobody labeled the internet; the text is its own answer key. This is the single most consequential idea of the last decade, because it unlocked training on essentially unlimited data.

Where it shows up for you

Every time an app finds 'similar photos', groups your expenses without being told the categories, or completes your sentence, some flavor of unsupervised or self-supervised learning is running. And when you use an LLM, you are talking to a self-supervised model that was later fine-tuned with supervised examples and human preferences: the modern stack layers all the paradigms.

In one glance