AI Foundations
Loss function
A mathematical score that tells a model how wrong its current predictions are.
Example
Training measures how far predicted house prices are from the known sale prices.
Why it matters
It gives the learning process feedback about which changes are helping.
You'll hear it
“The score improved, but it still makes the mistake customers care about.”
Can you control it?
Set by the team
Model builders choose or design the loss function before training. Users cannot change it in a finished hosted model.
Common questions
- Can different tasks need different loss functions?
- Yes. Predicting a price and choosing a category may need different ways of measuring mistakes.
- Is lower loss always better?
- It helps with the chosen target, but that target may miss other qualities or risks people care about.
- Is loss the same as accuracy?
- No. Accuracy counts correct choices; a loss can also reflect how large or confident mistakes are.