Think Together · Start with a Short Story
The girl shows the robot a black cat, a white cat, and a curly-haired dog. It cannot rely on a rule like “black means cat.”
Lay Out the Process
- PictureTurn it into numbers
- ModelCombine clues such as shape
- OutputScores for cat and dog
- CheckCompare with the actual category
An image classification model learns visual patterns related to categories through training.
It may combine clues such as shape and texture instead of looking at just one color. When a new picture goes in, the model calculates category scores and chooses a prediction according to its rules.
If training pictures are too similar, the model may rely on unreliable clues such as the background.
Now You Try
If every cat in the training pictures is on a sofa and every dog is on grass, what might the model learn by mistake?
Ready? Open this to see the explanation
It may treat the background as an important clue and misclassify a cat on grass. Use varied and suitable data to check for this kind of bias.
One Idea to Take Away
An image classification model learns visual patterns related to categories through training.
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.