Regularization

Regularization is a family of techniques that reduce overfitting by constraining how a model learns. L1 and L2 penalties discourage large weights, dropout randomly removes activations during training, early stopping limits optimization when validation performance stops improving, and data augmentation increases useful variation. Other methods smooth labels, inject noise, or restrict architecture. The goal is not simply to make a model smaller; it is to trade a little training-set fit for better performance on unseen data. The appropriate strength depends on dataset size, noise, model capacity, and task. Too much regularization causes underfitting, while too little allows memorization and fragile predictions.