Early Stopping

A Regularisation method for Neural Networks where we start the training with small weights and stop the learning when a minimum error on the test set is reached.

Early stopping has a prior towards linear models as the weights are still small and therefore in the linear range of the tanh function.