In the week 1 lab while calculating the loss of discriminator we call detach on the output from the generator because if we run update we don’t want the generator model weights to be updated, in a similar fashion when we are calculating the loss for the generator we don’t want the discriminator model weights to be updated so we should call for detach in generator loss function as well, but currently, we don’t can someone explain me why we don’t?
It turns out the situation is fundamentally asymmetric. This question comes up pretty frequently and here’s a thread which discusses it in some detail.
The explanation in the thread makes sense to me, Thank you.
But after reading the thread I feel, how could I ask such a dumb question?
No worries! As you can see from the previous threads, you are far from the first person to have this question. It comes up pretty frequently.