Think Together · Start with a Short Story
The girl hopes the robot will learn every kind of fruit at once. The robot pictures training as adjusting a dial in many small steps.
Lay Out the Process
- Make a first predictionSee how it performs at the start
- Calculate the errorNotice how far off it was
- Adjust a littleChange the parameters
- Evaluate againCheck performance on new data
Optimization often improves a model gradually through repeated parameter updates.
A step that is too large may skip a useful setting; a step that is too small may make training slow. The right training setup depends on the task.
A lower training error is only one signal. Also check validation data so the model does not simply memorize its practice examples.
Now You Try
The training score keeps rising, but the score on unseen pictures is falling. Should you keep training without checking?
Ready? Open this to see the explanation
No. The model may be overfitting. Check when to stop, the data, and the training setup.
One Idea to Take Away
Optimization often improves a model gradually through repeated parameter updates.
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.