How do you troubleshoot unbalanced training loss in a GAN where the discriminator becomes too strong