16. Help models generalize beyond the training set
Control overfitting with validation curves, weight decay, dropout, early stopping, data augmentation, and simpler models. The chapter gives you practical judgment for when a model has memorized instead of learned.