Kinds of learning: supervised, unsupervised, reinforcement

In short: There are three basic kinds: supervised learning (examples with the right answer), unsupervised learning (finding patterns without answers) and reinforcement learning (learning from reward). Mixed forms such as self-supervised learning exist too.

In more detail: Which kind fits depends on the data you have and what you want to achieve.

In detail

• Supervised: input + correct answer (label). Classification (spam yes/no) and regression (house price).
• Unsupervised: inputs only. Clustering (finding customer groups), dimensionality reduction (PCA), anomaly detection.
• Self-supervised: the answer is hidden in the data itself. Language models predict the next word, image models reconstruct masked image parts. This makes huge unlabelled data sets usable.
• Reinforcement: an agent tries actions in an environment and receives rewards. Examples: games (AlphaGo), robot control, fine-tuning chatbots (RLHF), see Reinforcement learning.
• Semi-supervised: few labels, lots of unlabelled data.
• In practice: first self-supervised pre-training, then supervised post-training, then reinforcement with feedback.

See also: What is machine learning?, Classification and regression