Analysis of improvement from trick of section 3.2

[carlito387]

Compare running times of new update VS original update on some dataset and for some loss functions. Convergence should be the same unless I didn't fully understand the trick @smolPixel can you confirm ?

[smolPixel]

Convergence is supposed to be faster with the new update, because the generator is directly battling the discriminator instead of trying to fit a function of the discriminator. Idk yet how much of a difference it will make but it's a standard trick in GAN so I imagine quite a bit