Overview

Every method in this unit so far has assumed labelled data: a target $y$ to predict from features $x$. This topic drops that assumption entirely. Unsupervised learning asks what structure a data set reveals about itself, with no labels to guide the search — grouping similar points together (clustering), finding a small set of latent factors that explain the data's variation (dimensionality reduction), or estimating the underlying probability distribution the data was drawn from (density estimation). The throughline across this topic's lessons is the latent variable: an unobserved quantity (a cluster identity, a low-dimensional coordinate, an independent source signal) that, if known, would make the data's structure obvious — and each method in this topic is, in essence, a different strategy for inferring it.