Questions

1. In a GAN, the generator and discriminator play the same roles as the encoder and decoder in an autoencoder. Is this correct?
2. Can a discriminator output values in the range (-1, 1)?
3. One of the problems with GANs is the premature convergence of the discriminator. It this correct?
4. In a Wasserstein GAN, is the critic (the discriminator) slower than the generator during the training phase?
5. Considering the previous question, what is the reason for the different speeds?
6. What is the value of the Jensen-Shannon divergence between U(-1, 0) and U(1, 2)?
7. What is the goal of the winner-takes-all strategy?
8. What's the purpose of the adjustment phase of the SOM training process?