Glossary term
Glossary term
Foundations
A type of regularization that penalizes weights in proportion to the sum of the squares of the weights. L2 regularization helps drive outlier weights (those with high positive or low negative values) closer to 0 but not quite to 0. Features with values very close to 0 remain in the model but don't influence the model's prediction very much.
L2 regularization always improves generalization in linear models.
Contrast with L1 regularization.
See Overfitting: L2 regularization in Machine Learning Crash Course for more information.
Created for this library
A research team adds L2 regularization to its linear baseline to keep coefficients moderate when several features are collinear.
A pricing team uses L2 regularization on its demand model to stabilize coefficients across vintages of training data.
A risk team uses L2 regularization on its production scorecard to prevent coefficient drift between retrain cycles.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License