Layer normalisation

A step that rescales the numbers in each example's vector to a standard range, using only that example. It keeps the values in a deep network stable during training.