Skip to content
    Flow-alerta.cab

    Early Stopping

    AI fundamentals & modelsEarly training discontinuationRegularization through training stop
    Editorially reviewed on: 1 source
    Your reading view

    The short answer · 17 words

    In short

    Technique that stops training if a monitored metric does not improve over a set period of time.

    Flow-alerta.cab (2026): Early Stopping. https://flow-alerta.cab/en/glossar/early-stopping. Editorially reviewed on: 10 October 2026.

    How much detail would you like?

    Practical

    Often useful for model training; choice of metric and patience must fit the course.

    Hypothetical example: Training is terminated after 25 epochs because the loss of validation no longer decreases during the established patience.

    Reduce training time, save cloud GPU costs and protect against over-adjusted prediction models.

    At a glance

    Core
    Often the loss of validation is monitored. Patience determines the number of epochs without sufficient improvement; min_delta sets the minimum change. The procedure can limit overfitting and unnecessary training, but it does not safely prevent them.
    Use
    Often useful for model training; choice of metric and patience must fit the course.
    Example
    Hypothetical example: Training is terminated after 25 epochs because the loss of validation no longer decreases during the established patience.
    Limits
    Patience parameters selected too small (training stops too early in the case of small fluctuations); no saving of the best checkpoint (Best Model Weights).

    Questions about this term

    What does Patience mean?

    The number of epochs without sufficient improvement before the stop.

    Does the best model always remain?

    Only with appropriate checkpointing or enabled restoration of the best weights; in Keras, restore_best_weights is False by default.

    All details and background

    Often the loss of validation is monitored. Patience determines the number of epochs without sufficient improvement; min_delta sets the minimum change. The procedure can limit overfitting and unnecessary training, but it does not safely prevent them.

    ELI5 – Explained simply

    As with cake baking: you take the cake out of the oven as soon as it is finished, instead of waiting strictly after the clock until it burns.

    Why it matters in practice

    Often useful for model training; choice of metric and patience must fit the course.

    Example

    Hypothetical example: Training is terminated after 25 epochs because the loss of validation no longer decreases during the established patience.

    Pitfalls and limits

    • Patience parameters selected too small (training stops too early in the case of small fluctuations); no saving of the best checkpoint (Best Model Weights).

    Origin & history

    Early stopping is an established training control; modern frameworks provide it as a callback.

    Use in mid-sized companies

    • Reduce training time, save cloud GPU costs and protect against over-adjusted prediction models.

    Related terms

    • Overfitting
    • Regularisation
    • Hyperparameter tuning
    • Checkpointing
    • Validation loss

    Matching Flow-alerta service

    TrainingAI consulting: Examine the suitability, data and prerequisites of a possible use case on the topic of “Early Stopping Training”

    Matching tool / use case

    View use casesUse Case: Control training based on a validation indicator / Tool example: Keras EarlyStopping

    Sources

    Quellenprüfung: 10.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.

    Editorially reviewed on:

    Next step

    Put knowledge into practice

    Discuss your projectView use cases