Overfitting
과적합
When a model fits its training data well but fails on new data.
It is like a student who memorized the practice test: perfect on those questions, lost when they change.
Detection is simple: evaluate on data the model never trained on, a validation set.
Remedies include more data, a smaller model, stopping training earlier, and regularization.
- SignalTraining score high, validation score low.
- The oppositeToo simple to fit either is underfitting.