Deep learning, and rules versus learning
Deep learning is a kind of machine learning. It uses a neural network with many layers, loosely inspired by the brain; the word loosely matters, because it is a mathematical model and not a small brain. Deep learning is used for recognising faces and voices and for translation. Now compare two kinds of rules. In the program below, a person decided that 50 is the pass mark and wrote it into the code. This is a fixed rule: it never changes unless a person changes it. A learned rule would be found from data, for example from past results: the model would look at the study hours of earlier pupils and find where passing starts. A learned rule improves with more and better examples, but it needs training, and it can still be wrong.
1score = 62 2if score >= 50: 3 print("Pass") 4else: 5 print("Fail")
Pass
| Fixed rule | Learned model | |
|---|---|---|
| Who makes the rule | a person | the model, from data |
| Does it change | only if a person edits it | can improve with more examples |
| Needs training | no | yes |
| A | B | |
|---|---|---|
| 1 | Study hours | Result |
| 2 | 1 | fail |
| 3 | 2 | fail |
| 4 | 3 | fail |
| 5 | 4 | pass |
| 6 | 5 | pass |
| 7 | 6 | pass |
Write one fixed rule and one rule that a model could learn from data, both for sorting school bags into heavy and light. Say who chooses the border in each case.