Training a model, explained simply
Machine learning is a way of making a computer good at a task by showing it examples instead of writing step-by-step instructions. The computer looks for patterns in the examples and builds a model.
A model is the result of training: a kind of summary of the patterns it found. When you give the model something new, it uses those patterns to make a guess, called a prediction.
The model does not understand things the way you do. It has only noticed which features tended to go with which labels. More varied, well-labelled examples usually give it a better sense of the pattern.
Example: Learning what a berry looks like
- You show the computer 20 berries and 20 crystals, all labelled.
- Training: it notices that berries are usually round and coloured, crystals pointy and clear.
- The model: a stored pattern it can reuse.
- Prediction: shown a new round, red object, it guesses "berry".
Try this
- Play "guess my rule": think of a secret rule, show examples that follow it, and see how many examples a friend needs to guess it.
- List three apps your family uses and guess what examples each one might have learned from.
- Talk about how you learnt to recognise a dog. Did anyone give you rules, or did you see lots of dogs?