WS standardizes the weights in convolutional layers to clean the reduction landscape by lessening the Lipschitz constants of your decline as well as the gradients; BCN brings together batch and channel normalizations and leverages believed statistics with the activations in convolutional layers to help keep networks clear of elimination singularities. https://bt-ch-channel70245.theobloggers.com/29466002/bt-ch-tv-channel-an-overview