Supervised learning uses data where the correct answer is already known.
The model learns the relationship between inputs and their matching outputs.
For example, emails already labelled as spam or not spam.
It learns from those examples and then labels new emails on its own.
Regression and classification are the two big supervised tasks.
Supervised learning is learning from worked examples.
Each example arrives with the right answer attached to it.
The model's job is to copy the relationship, not the examples themselves.
It is the most common kind of machine learning in real products.
It is also the most expensive, because someone had to produce those answers.
If the answers came from another system, you inherit that system's mistakes.
Two tasks cover almost all of it: predicting a number or predicting a category.
The model never learns why an answer is right, only that it is.
So it will happily latch onto a shortcut that happens to work in your data.
In one famous case a model spotted rulers in photographs of skin cancer.
The rulers were there because doctors measured the serious cases.
The model learned rulers, not cancer, and nobody noticed for a while.
Supervised learning is powerful precisely because it is narrow.
It answers the question you labelled and no other question.
Change what you want to predict and you usually need new labels.
One question on this card. Every option explains itself once you pick, right or wrong.
55 concepts in this module, then a graded test with every answer explained.