Module 4 of 12

The Three Types of Machine Learning

About 9 minute read

Machine learning splits into three broad families, distinguished by what kind of data they learn from and what kind of feedback guides them.

Machine LearningSupervised
Learns from labeled examples (input → correct answer)
e.g. spam filtering
Unsupervised
Finds structure in unlabeled data on its own
e.g. customer segments
Reinforcement
Learns by trial and error, guided by rewards
e.g. game-playing agents
The three broad families of machine learning.

Supervised learning

Supervised learning is what Module 3 described: you have labeled examples (input, correct answer), and the model learns to map inputs to outputs. It further splits into two common problem types:

  • Regression — predicting a number (house price, temperature tomorrow). Module 5 covers this with linear regression.
  • Classification — predicting a category (spam/not spam, cat/dog/horse). Module 6 covers this with k-Nearest Neighbors.

Most of the AI you interact with daily — spam filters, recommendation systems, fraud detection, voice recognition — is supervised learning.

Unsupervised learning

Here, there are no labels at all — just raw data — and the goal is to find structure in it. Common tasks include:

  • Clustering — grouping similar items together (e.g. segmenting customers by purchasing behavior without being told the segments in advance).
  • Dimensionality reduction — compressing data down to its most important patterns while losing as little information as possible (useful for visualizing high-dimensional data, or as a preprocessing step for other models).
  • Anomaly detection — flagging data points that don't fit the normal pattern (e.g. unusual credit card activity).

Reinforcement learning

Reinforcement learning (RL) is different from both: an agent takes actions in an environment and receives rewards or penalties, learning through trial and error which actions lead to good long-term outcomes — much like training a dog with treats. There's no fixed dataset of "correct answers"; the agent generates its own experience by acting.

RL is behind game-playing systems (like agents that mastered Go and chess through self-play), robotics control, and is also used to fine-tune the behavior of some large language models based on human feedback (often abbreviated RLHF).

Key idea: These categories aren't mutually exclusive in practice — many real systems combine techniques from more than one family. But knowing the three lenses (learn from labels, learn from structure, learn from trial and reward) will help you make sense of almost any AI system you read about.
Check your understanding
1. A system that groups customers into segments without ever being told what the segments should be is an example of:
2. A game-playing agent that improves by receiving rewards for winning moves and penalties for losing ones is using:
3. Predicting a house's sale price from its features is an example of which supervised learning task?
Up next · Module 5 of 12Your First Model: Linear Regression Previous: How Machines Learn From Data
Did you like the lesson? 😆👍
Consider a donation to support our work:
See donation links