Definition
A technique that normalizes the inputs *across* the features (neurons) within a single layer of a neural network for a single training example, stabilizing training.
In short: A technique that normalizes the inputs *across* the features (neurons) within a single layer of a neural network for a single training example, stabilizing training.