The Big Idea: Learning from Examples, Not Instructions
In traditional programming, a person writes exact instructions for the computer: "if the fruit is red and round, call it an apple." But for many real problems — recognising faces, understanding speech, spotting spam — nobody can write down all the rules. There are simply too many, and many of them we cannot even describe in words.
Machine learning flips the approach: instead of giving the computer rules, we give it lots of examples and let it discover the rules on its own. The computer studies the examples, finds patterns that repeat, and stores a compressed summary of those patterns. That summary is called a model, and it can then make predictions about things it has never seen before.
By the end of this lesson you will be able to:
- Explain the difference between programming with rules and learning from examples
- Define data, features, and labels, and identify them in everyday situations
- Describe how repeated examples reveal patterns, and what a model really is
- Tell apart the training phase and the using phase (inference)
- Explain why more and better data usually means a better model — and what "garbage in, garbage out" means
1. Two Ways to Make a Computer Smart
Imagine you want a computer to sort fruit into apples and oranges. There are two very different strategies:
A human writes the rules by hand:
IF color is red AND shape is round
THEN it is an apple
IF color is orange AND skin is bumpy
THEN it is an orangeWorks for simple problems, but breaks the moment reality gets messy — green apples exist, some oranges are yellowish, lighting changes colors in photos.