You don't /need/ a source of randomness for GAN inference; it just helps with some applications. The basic idea is Network G produces some output, and Network D

You don't /need/ a source of randomness for GAN inference; it just helps with some applications. The basic idea is Network G produces some output, and Network D tries to classify it...