What is early stopping in deep learning?

Early stopping is a training technique that stops a model when its performance on validation data stops improving.

It is mainly used to reduce overfitting and avoid unnecessary training.

How it works

During training, the model is evaluated on a validation set after each epoch or at regular intervals.

If the chosen validation metric does not improve for a certain number of checks, training is stopped.

That waiting period is often called patience.

For example:

Train → Validate → Improve? → Continue

If validation performance stops improving for several epochs:

Stop training → Restore best model checkpoint

Why is early stopping useful?

It can help:

• Reduce overfitting.
• Save training time and compute.
• Preserve the model state that performed best on validation data.

Important point

Early stopping should be based on validation performance, not training performance alone.

Training loss may continue to decrease even while validation performance begins to get worse.

For the related problem, see “What is overfitting in deep learning?”.

In short, early stopping ends training when validation performance stops improving, helping prevent overfitting and unnecessary computation.