Supervised vs Unsupervised Learning — The Basic Split

Why it matters

Choosing the wrong paradigm for a problem wastes effort. Understanding when each applies is fundamental ML knowledge.

The architecture

Supervised: classification (discrete labels) and regression (continuous).

Unsupervised: clustering (k-means), dimensionality reduction (PCA), density estimation.

How it works end to end

Self-supervised: modern middle ground. Extract labels from data automatically. Next-word prediction, masked-language modeling. Enables LLM pretraining without human labels.

Semi-supervised: mix labeled + unlabeled. Useful when labels expensive.

Reinforcement learning: separate paradigm. Learn from reward signal.