AI Glossary Term
What does “Overfitting” mean?
Definition
When a model memorizes training data and performs poorly on unseen inputs.
How Overfitting works
Example
A model memorizes that a particular photo background always appears with the label 'wolf', instead of learning what wolves look like. It scores perfectly on training pictures and fails on new ones.
Why it matters
Overfit models look excellent in development and disappoint in use. Held-out testing is the main safeguard.
Common questions
How do I detect overfitting?
Training performance is much better than validation or test performance.
How do I prevent it?
More data, simpler models, regularization such as dropout, data augmentation, and early stopping.
What is underfitting?
The opposite: the model is too simple or undertrained and performs poorly even on training data.