Introduction
Think about the different ways people learn:
- A student with a teacher who marks every answer right or wrong.
- A tourist in a new city who walks around and slowly notices groups of similar neighbourhoods — nobody tells them the names.
- A child learning to ride a bike who tries, falls, adjusts and gets better through rewards (staying up) and penalties (falling).
Machine learning systems learn in the same different ways. The biggest difference between them is how much "supervision" they get — that is, whether the training examples come with the correct answers or not.
The four main families are:
| Type | Gets correct answers? | Everyday analogy |
|---|---|---|
| Supervised | yes, for every example | learning with a teacher |
| Unsupervised | no | exploring and finding groups yourself |
| Semi-/self-supervised | for a few, or creates its own | a teacher for some lessons, self-study for the rest |
| Reinforcement | only rewards and penalties | learning a game by playing it |