Batch Normalization
Normalizing layer activations to stabilize and speed up training.
Batch normalization rescales the activations within each mini-batch so they have consistent statistics, which smooths optimization and allows higher learning rates. It was a key enabler of training very deep vision networks.