Early stopping
Also called Early Stopping
Stop training the moment the model stops getting better on held-out data, before it starts memorizing.
Think of it like
Taking the cake out when the toothpick comes clean, not leaving it in until the edges burn.
Example
A run keeps improving on the training data but its validation loss ticks up at epoch 12 — early stopping halts there and keeps the epoch-11 weights.
How it actually works
You watch a validation metric each epoch and stop when it has not improved for a set number of rounds (the patience), keeping the best checkpoint seen. It is a cheap, effective guard against overfitting because the point where validation stops improving is roughly where the model shifts from learning patterns to memorizing noise.
For product teams
A simple way to avoid wasting compute and shipping an over-trained model that looks great on paper and worse in the wild.
For engineers
Monitor a validation metric with a patience counter; restore the best checkpoint when it plateaus or regresses.
Related
- Validation set — The signal it watches to decide when to quit.
- Overfitting — The thing it is trying to prevent.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome