Skip Norm is a new and flexible building block for Deep Learning that serves as a skip connection and normalization block simultaneously. It outperforms BatchNorm, LayerNorm, and GroupNorm when used in generative models such as Autoencoders.
Skip Norm is a new and flexible building block for Deep Learning that serves as a skip connection and normalization block simultaneously. It outperforms BatchNorm, LayerNorm, and GroupNorm when used in generative models such as Autoencoders.