Early Stopping

Always a good idea!

Stop training the model when accuracy on the validation set decreases.
Or train for a long time, but always keep track of the model snapshot that worked best on val.
Pasted image 20241202170102.png