Rocket wants the sorting model to learn from examples. He dumps photos into a bin.
"Learn, sorting model, learn!" he says. Nothing happens.
Raven picks up a photo. "What do you notice? Nothing says which label it gets."
Nova hovers over the bin. "Here is a hint," she says. "How would anyone learn from examples with no answers?"
Raven grabs a stack of sticky notes.
To learn from examples, each example needs the right answer.
The right answer attached to an example is called a label.
A rock photo with a note that says rock is a labeled example.
A big set of labeled examples is called training data.
| Example | Label | Labeled? |
|---|---|---|
| Photo of a green fern with leaves | plant | yes |
| Photo of a hammer with a handle | tool | yes |
| Photo of a lumpy gray stone | (no label) | no |
Training data usually comes from people.
People collect the examples. People often add the labels too.
So people shape what a learning model learns.
That is a big job, and it matters a lot!
| Statement | True or false? |
|---|---|
| A label is the right answer for an example. | ? |
| Training data is usually supplied by people. | ? |
| A model can learn from photos with no labels at all, in our lab. | ? |
| People shape what a learning model learns. | ? |
Great labeling, explorer! Tomorrow we look for patterns in the labeled photos.