Why to use activation units/functions (non-linearity) inside convolution neural networks (CNNs)

Open in new window