Glossary term
Glossary term
Foundations
A method for regularization that involves ending training before training loss finishes decreasing. In early stopping, you intentionally stop training the model when the loss on a validation dataset starts to increase; that is, when generalization performance worsens.
Click the icon for additional notes.
Contrast with early exit.
Created for this library
An ML platform team enables early stopping on validation loss so production training jobs do not waste compute past the point of overfitting.
A computer vision team uses early stopping with a patience of five epochs to terminate training as soon as the validation curve flattens.
A churn modeling team uses early stopping with a held-out month of data so the production model is the best version on unseen recent behavior.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License