Unsupervised learning

Find structure in data without labels.

Clustering

Group similar things together. "What customer segments naturally exist in our data?"

Dimensionality reduction

Compress thousands of variables down to a handful that capture most of the signal. The intuition behind “embeddings.”

Anomaly detection

Flag observations that don’t look like everything else. Fraud, safety incidents, manufacturing defects.

Applied frame

Name the input, target, feedback signal, and evaluation split before choosing a model family.

Diagram cue

The diagram traces data into a learned function and back out through evaluation.