I don’t understand machine learning. I’ve run your code also on Intel XEON ® E5–2673 v4 (Broadwell) processor, and I received also the same results — nothing more than big white spots. But I managed to buy VPS on Azure with GPU NVIDIA Tesla K80, and to my big surprise some digits appears as soon as at 100 steps and improve constantly. Below I’ve pasted image of 1000 step.
@Rowel Atienza If possible could you explain, why your code works only on GPU, and it doesn’t work on CPU?
What is so different between these architectures? It seems, that on CPU neural networks fall into some local minimum from 100 steps above, but on GPU this problem doesn’t appear and GAN is improving all the time. This is very surprising for me. Any explanation appreciated :) And thanks for amazing article and code once again :)

