Generalization Introduction
Regularization
Encourage models to fit patterns that carry beyond training data.
Before this
The intuition
Regularization constrains or penalizes a model to discourage excessive sensitivity to training examples. Its strength is a modeling choice, usually selected using validation data.
A concrete example
Ridge regression adds a penalty proportional to the sum of squared weights to the fitting objective. Increasing its strength encourages smaller weights.
What to remember
More regularization is not always better. Too much can prevent a model from fitting useful structure.
Check your understanding
Which data should guide the choice of penalty strength?
Reveal explanation
Validation data, keeping the final test set reserved for evaluation.