Early stopping

How it is used

Used to limit overfitting and unnecessary training. A patience setting may allow several checks without improvement.

Example sentences

1. The early stopping rule ended training after the validation loss failed to improve for five epochs.

   The stopping rule responds to validation performance rather than a fixed total epoch count alone.

2. We restored the best checkpoint after early stopping instead of keeping the final weights.

   The last training state may be worse than the earlier state that achieved the best validation score.

Common collocations

• early stopping patience
• enable early stopping

Related terms

• overfittingA problem in which a model learns its training data too closely, including noise and accidents, so it scores well on that data but poorly on new data.
• validation setA part of the data kept out of training and used during development to compare models, tune settings and decide when to stop training.

More terms: Machine Learning and Data Processing

• artificial intelligence
• machine learning
• model
• dataset
• training data
• label
• train
• supervised learning
• neural network
• deep learning
• loss function
• test set