tensorflow-generative-model-collections icon indicating copy to clipboard operation
tensorflow-generative-model-collections copied to clipboard

Published 20 hours ago • verified publisher icon hwalsuklee

Fix ACGAN training

Open yang-song opened this issue 7 years ago • 1 comments

The ACGAN training procedure is not optimal

yang-song avatar Mar 18 '18 07:03 yang-song

I think this is correct and makes training stable.

By the article,

$$L_s=E[logP(S=real|X_{real})]+E[logP(S=fake|X_{fake})]$$ $$L_c=E[logP(C=c|X_{real})]+E[logP(C=c|X_{fake})]$$ $$D = argmax(L_s+L_c)$$ $$G = argmax(L_c-L_s)$$

which means, you don't need to train Q separately. Also, Q shouldn't optimize variables of G.

ElApseR avatar Feb 06 '19 08:02 ElApseR