Overfitting
When a model memorizes its training data instead of learning the pattern, and flops on anything new.
Think of it like
A student who memorizes the practice exam word for word, then panics when the real questions are reworded.
Example
A model that scores 99% on training data but 60% on fresh data is overfit — it learned the answers, not the lesson.
How it actually works
Overfitting happens when a model has enough capacity to fit noise and quirks specific to the training set. It looks great in training and fails to generalize. Remedies include more data, regularization like dropout, and stopping training before it starts memorizing.
For product teams
It’s why a model that “aced testing” can still disappoint in production — check performance on held-out data.
For engineers
Low training loss but a high generalization gap; the model fits noise rather than signal.
Related
- Underfitting — The opposite is underfitting.
- Generalization — The goal it violates is generalization.
- Dropout — Countered by dropout.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome