Why don't we detach the discriminator?

[NikRan]

Why don’t we detach the discriminator when training the generator like we detach the generator when training the discriminator?

[gent.spah]

Further on this place of the forum you could find this explaining it:

Why should we detach the discriminators input ?!

in the first programming assignment of week 4, in the training part, it has been commented that “remember to detach the generator” and I detached the fake variable before giving it to the discriminator and it worked. but it doesn’t make sense to me because when we do backpropagation to update the generator, the computational graph starts from the loss values then goes back through the discriminator without updating its parameters, and from the discriminators input it finds its way back to the g…