Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Unsupervised learning

All methods discussed so far are supervised: every training example comes with a known target value yiy_i, and the model is trained to reproduce it. In practice, however, labels are often expensive or impossible to obtain at scale — most of the world’s data is unlabeled.

Unsupervised learning works with the input features X\boldsymbol{X} alone, with no target to predict, and instead aims to discover structure within the data itself: groups of similar samples (clustering), a lower-dimensional representation that retains the most informative directions of variation (dimensionality reduction), or a compact learned encoding of the data (representation learning). These techniques are valuable on their own, for exploration and visualization, and as a preprocessing step that can make downstream supervised models more effective, especially when labeled data is limited. The next three notebooks cover three core unsupervised methods: K-means clustering, principal component analysis, and autoencoders.