Complex models memorize. Regularization penalizes complexity, pushing the model toward simpler solutions that generalize better. It is one of the most reliable tools against overfitting.

L2 regularization adds a penalty proportional to the square of weights, shrinking them toward zero. L1 encourages sparsity by penalizing absolute values. Dropout randomly disables neurons during training, forcing the network to be robust. Early stopping halts training when validation loss stops improving.

Common regularization methods

• L1 and L2 weight penalties
• Dropout
• Early stopping
• Data augmentation
• Batch normalization

Too much regularization causes underfitting. The model becomes too simple to capture real patterns. Tuning the strength is part of the craft, and cross-validation usually guides the choice.