Another week of programming has passed and I feel so much comfortable with
GANs already. I spent the entire week debugging Kris'
GAN code, and implementing the O'Reilly test on the
Though progress went a lot slower than I expected, now the core infrastructure is fully functional, and the only task that remains is to find the appropriate hyper-parameters for the
GAN code, which would be my goal for the next couple of days. I'll also have to consult with Marcus and Mikhail regarding changing the current
GAN optimizer API, though this would not be of much importance if we are able to obtain good results with the existing code.
I have a lot to cover up to stay on track of getting the
DCGAN implementation up and functional within the next two weeks. Hopefully, the current
GAN code wouldn't take much longer to get merged, and we could leverage the experience gained while implementing
DCGAN which follows closely from the standard implementation.
Generated by 1.8.13