In the “Model Collapse” video, what does the presenter mean by the following: “Take a discriminator that has learned to be good at identifying which handwritten digits are fakes …”
What is a “fake” handwritten digit???
One created by your generator function. This is just a confusion of language: it’s not “hand-written”, but the point is you are training your generator to create synthetic images that look like hand-written digits. That’s the whole point of this exercise, right?
Thanks for the reply Paul. For some reason, I’m having a mental block with GANs. I’ve taken all the deep learning courses which I found very intuitive, but GANs make no sense to me. I’ll just keep plugging along, and hopefully everything will click into place …