DCGAN is initialized with random weights, so a random code plugged into the network would create a totally random impression. On the other hand, when you may think, the network has millions of parameters that we can tweak, as well as goal is to find a setting of these https://ambiq.ai/models/heart-arrhythmia-classification/