A technique used to normalize the inputs of each layer in a deep neural network, improving its performance and robustness.