As we can see here the image generated at the 71th epoch is more clear(clean) than the image generated at the 440th step. I was thinking with more epochs the generation should be enhanced ,but reverse is happening. What is the reason behind such a strange behaviour.
It’s been a while since I took these courses, but I’m pretty sure that Prof Zhou discusses this in some detail in the lectures. It turns out that convergence is never guaranteed. There is always a possibility that the training of the discriminator and the generator gets unbalanced and one wins out. Even if the training is going well at a particular epoch, that doesn’t mean things can’t turn sour later.