Hello Model
← Training basics

Lesson 4 of 8

Overfitting and underfitting

In plain words: Overfitting is memorising the training data; underfitting is not learning enough from it.
  • Overfitting: training score keeps improving while validation score gets worse.
  • Fixes: more data, Data augmentation, simpler model, regularisation, early stopping.
  • Underfitting: both training and validation scores are poor.
  • Fixes: a more powerful model, better features, train longer.
  • Plot training vs validation scores per epoch — the gap tells you which problem you have.
Tip Early stopping (stop when validation stops improving) is the easiest overfitting fix.