Think Together · Start with a Short Story
The girl brings out an answer card, and the robot compares its guess with the answer. “Knowing it was wrong” here does not mean feeling regret like a person.
Lay Out the Process
- ExampleA picture of a cat
- PredictionLeans toward “dog”
- Target labelCat
- LossMeasures the difference
During training, a system can calculate error using a target and a loss function.
In classification with labels, the target class is compared with the predicted distribution to create a signal for optimization. This is a numerical calculation, not the model privately admitting a mistake.
If the label itself is wrong, the feedback will be wrong too, so checking the data matters.
Now You Try
If a cat is mislabeled as a dog, will the training system automatically know the teacher made a mistake?
Ready? Open this to see the explanation
Not necessarily. It may adjust itself using the wrong label. The data needs quality checks and independent validation.
One Idea to Take Away
During training, a system can calculate error using a target and a loss function.
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.