Think Together · Start with a Short Story
The robot guesses a fruit incorrectly for the first time, and the girl tells it the correct category. How does the computing system adjust next?
Lay Out the Process
- PredictGive the current answer
- CompareCalculate the difference from the target
- AdjustUpdate parameters
- RepeatPractice with more examples
Training often improves performance by calculating the difference between predictions and targets, then updating parameters.
The numerical error is called loss. It gives the optimization process a direction it can calculate. An adjustment does not simply fix a typo; it changes parameters that can affect many inputs.
Each update may not make every example better, so performance needs to be checked overall and through independent evaluation.
Now You Try
After correcting one cat picture, can you guarantee every other picture will now be classified more accurately?
Ready? Open this to see the explanation
No. A parameter change may improve some cases and affect others, so evaluate with varied data.
One Idea to Take Away
Training often improves performance by calculating the difference between predictions and targets, then updating parameters.
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.