A technique used to speed up the training process of neural networks by normalizing the input to each layer.