Glossary term
Glossary term
Foundations
A number that specifies the relative importance of regularization during training. Raising the regularization rate reduces overfitting but may reduce the model's predictive power. Conversely, reducing or omitting the regularization rate increases overfitting.
Click the icon to see the math.
See Overfitting: L2 regularization in Machine Learning Crash Course for more information.
Created for this library
An ML team tunes the regularization rate on a held-out set per model release to balance bias and variance.
A risk team locks the regularization rate at release time so production retraining stays reproducible across vintages.
A research team sweeps the regularization rate to find a stable region on small datasets before locking the production value.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License