A Regularization technique where we have a validation data set, which we use for Cross-validation. We stop training, when the validation error (which estimates future prediction error) is lowest (before it starts increasing due to Overfitting, thus the name early stopping).