unsupervised learning
/ un-soo-per-VIZED LER-ning /
Imagine you tip out a giant box of mixed buttons on a table and, without anyone telling you the categories, you start grouping them — these are round and red, those are square and metal. Nobody handed you the right answers; you found structure just by looking. That is unsupervised learning: a computer studies data that carries no labels and tries to discover the hidden patterns, groupings, or shape inside it on its own.
More precisely, unsupervised learning works with inputs only — no target answer attached to each example. The two classic jobs are clustering (sorting examples into natural groups, like customers who buy similar things) and dimensionality reduction (boiling many measurements down to a few that capture the gist, so the data is easier to see and store). Because there is no answer key, success is judged indirectly — does the grouping look sensible, does it help a downstream task — rather than by a simple score.
It matters because labeled data is expensive and rare, while raw, unlabeled data is everywhere. Unsupervised methods let you explore a fresh dataset, spot anomalies, or compress information before any human has tagged a thing. The honest catch: with no answer key, results can be ambiguous. Different methods carve the same data into different groups, and there is rarely one provably correct answer — you, the human, still have to decide whether the structure it found is meaningful.
An online shop has no labels saying "bargain hunter" or "luxury buyer." It feeds each shopper's purchase history into a clustering algorithm, which groups them into a handful of natural segments. The marketing team then reads the groups and names them — the algorithm found the clusters; the humans gave them meaning.
Clustering finds the groups; people decide what they mean.
"Unsupervised" does not mean "no human effort" — you still choose the method, set its knobs, and judge whether the patterns are real. And there is no single right answer, so two reasonable runs can disagree.