1 paper · 1 filter
Florian Schäfer, Hongkai Zheng, Anima Anandkumar
To improve the stability of GAN training we need to understand why they can produce realistic samples. Presently, this is attributed to properties of the divergence obtained under…