The Big Idea: One Goal, Three Ways of Learning
In the previous lessons you saw that machine learning (ML) means letting a computer learn patterns from data instead of following hand-written rules. But not all learning problems look the same. Sometimes we have a teacher who shows the computer the right answers. Sometimes we have a pile of data and no answers at all. And sometimes there is no data to start with — the machine must learn by trying things and seeing what happens.
These three situations give us the three classic types of machine learning: supervised learning, unsupervised learning, and reinforcement learning. Almost every AI system you meet — spam filters, music recommendations, game-playing bots — belongs to one of these families. Knowing which is which tells you a lot about how the system was built and what data it needed.
Learning Goals
- Explain what supervised learning is and recognize its two flavors: classification and regression.
- Explain what unsupervised learning is and describe clustering with everyday examples.
- Describe reinforcement learning using the words agent, action, environment, and reward.
- Compare the three types by the data they need, their goal, and typical uses.
- Recognize the names semi-supervised and self-supervised learning when you hear them.
The Family Tree of Machine Learning
The three main branches of machine learning and their best-known sub-types.