Overfitting
Overfitting is when a model fits the noise and detail of training data too closely, giving low training error but high generalization error.
Purpose
To identify and prevent overfitting and improve model generalization.
Mechanism
When model complexity is too high, the model memorizes training data instead of learning the underlying pattern. Detection: training error far below validation error. Defenses: regularization, dropout, early stopping, data augmentation, simpler models, and ensembles.
Use cases
- Model generalization
- Deep learning
- Model selection