Hello,
For the GAN algorithm, there are two different versions, one in the original slides and one on the annotated ones, but both seem wrong to me. I saw in another post that the original ones a supposed to be correct, but in these, we are taking the gradient wrt the loss function of the generator to update the discriminator. Shouldn't it be wrt to the loss function of the discriminator? And if not can explain why, please?
Hi, thanks for noticing it!
The annotated version is almost correct (if the two losses are swapped), but anyway it's not consistent with the text above the algorithm, sorry about the confusion!
It should be as in the figure below:
GAN algorithm
Hello,
For the GAN algorithm, there are two different versions, one in the original slides and one on the annotated ones, but both seem wrong to me. I saw in another post that the original ones a supposed to be correct, but in these, we are taking the gradient wrt the loss function of the generator to update the discriminator. Shouldn't it be wrt to the loss function of the discriminator? And if not can explain why, please?
5
I was wondering the same thing so any help would be greatly appreciated.
Hi, thanks for noticing it!
The annotated version is almost correct (if the two losses are swapped), but anyway it's not consistent with the text above the algorithm, sorry about the confusion!
It should be as in the figure below:
6
Can you please explain what is the aim of the total objective function if then we consider a different loss function for D and G respectively?
thanks
1
Add comment