The examples set the goal

Consider one specific task: teaching a network to sort photographs into cats and dogs.

Each training photo comes with a , such as "cat" or "dog." A person or another process supplied it before training began.

The network makes a prediction and compares it with the label. The difference produces the error that guides the weight updates.

The labels tell the network which final distinction matters. They do not tell the hidden layers which edges, shapes, or other clues to use along the way.

People set the task through the examples and labels. Training searches for weights that perform that task well.