[2,1,4,4] is a vector with small numbers I chose so that calculation does not produce huge numbers. It’s just a random example to show the transposed convolution matrix calculation in that the weight layout is transposed from that of the convolution matrix. So, the actual weight values are irrevant here.
Hope that clarifies!
That’s a good question. It’s actually a convention to define such case as an infinity for KL divergence.
With GAN, we train the generator and the discriminator in turns.
When the generator is trained, the discriminator is used as a classifier and as such it is set to non-trainable.
When the discriminator is trained, it needs to be trainable and trained with real data…