Batch normalization layer (Ioffe and Szegedy, 2014). Normalize the activations of the previous layer at each batch, i.e. applies a transformation that maintains the mean activation close to 0 and the activation standard deviation close to 1. Arguments. axis: Integer, the axis that should be normalized (typically the features axis).