Think Together · Start with a Short Story
The girl wants to tell cats from dogs. One way is to write rules by hand; another is to train a model with many examples.
Lay Out the Process
- Rule-based programA person writes explicit decisions
- Learning systemData helps adjust the model
- UseGive it a new photo to predict
Traditional rule-based programs follow rules people write explicitly. Machine learning adjusts a model’s internal parameters from training data.
For example, fur colors vary too much for a person to write every possible rule for cats and dogs. A model can learn related patterns from varied examples instead.
A learning system still runs as a program. Real applications also often combine rules and models.
Now You Try
Does giving a model more cat and dog examples mean someone writes a separate “if” rule for every photo?
Ready? Open this to see the explanation
No. Training usually adjusts shared model parameters so the same calculations can handle many inputs. But the model can still learn poorly or memorize too much.
One Idea to Take Away
Traditional rule-based programs follow rules people write explicitly. Machine learning adjusts a model’s internal parameters from training data.
Try covering the diagram above and explaining each step to someone in your family. If one step is hard to explain, return to that card and look again.