Definition
A technique to stabilize and accelerate deep network training by normalizing the input mean and variance for each mini-batch across a layer.
In short: A technique to stabilize and accelerate deep network training by normalizing the input mean and variance for each mini-batch across a layer.