AI Foundations
Label
The correct answer or target attached to an example in supervised learning.
Example
Reviewers mark customer messages as positive or negative so a model can learn from them.
Why it matters
It gives learning and testing a clear target instead of leaving the expected answer unstated.
You'll hear it
“The reviewers disagree about which label this message should get.”
Can you control it?
Set by the team
Data teams define labeling rules and people or automated systems apply them. End users normally cannot change the labels used to train a finished model.
Common questions
- Can an example have several labels?
- Yes. Some tasks allow several descriptions, such as marking a picture as both outdoors and crowded.
- Who creates data labels?
- Human reviewers, existing records, sensors, rules, or other automated systems can produce labels.
- What happens when labels are wrong?
- The model may learn the wrong pattern and perform poorly or unfairly on new examples.