Hello there !
I have some troubles with the formulation of the parameter updates to train the generator and the discriminator.
Why do we update the discriminator with the generator objective function gradient, and vice versa ?
I might be missing some details about the min max optimization or the order of updates but I don't get it...
Thanks in advance for your explanations :)
In the original slides, for the discriminator update, we are taking the gradient of the generator's loss function wrt the discriminator parameter. is this the correct algorithm?
Alternating GAN algorithm
Hello there !
I have some troubles with the formulation of the parameter updates to train the generator and the discriminator.
Why do we update the discriminator with the generator objective function gradient, and vice versa ?
I might be missing some details about the min max optimization or the order of updates but I don't get it...
Thanks in advance for your explanations :)
there was a typo in the annotated slides. can you download the new original slides PDF?
2
In the original slides, for the discriminator update, we are taking the gradient of the generator's loss function wrt the discriminator parameter. is this the correct algorithm?
2
Thank you !
1
Very sorry for the typo should be:
1
Add comment