Computing Library › Machine Learning
Machine Learning

Unsupervised Learning

Unsupervised learning finds structure in data that carries no labels: groups, low-dimensional coordinates, or a density.

Learning without labels

Unsupervised learning receives only inputs x, with no target y. Instead of predicting a known answer it uncovers latent structure: which points cluster together, which directions carry most variation, or what probability density generated the data.

Main families

Kronos motion — density profile

Why it is harder to evaluate

With no ground-truth label, there is no single accuracy number. Success is judged by internal criteria (compactness, reconstruction error), by stability under resampling, or by whether the discovered structure helps a downstream task. This makes validation a matter of judgement as much as arithmetic.

A common pattern is to use unsupervised dimensionality reduction as a preprocessing step: project noisy high-dimensional inputs into a compact space, then run a supervised model on the compressed coordinates. This reduces variance and can defeat the curse of dimensionality.

Where it fits

Unsupervised methods are the right tool when labeling is expensive or impossible, when you are exploring a new dataset, or when you want to compress signals before modeling. In fusion diagnostics, clustering raw sensor traces can reveal distinct operating regimes that no one labeled in advance, guiding later supervised analysis.