GAN improvements that lead towards finding global minima for any given distribution mapping. GANs are versatile but still suffer from mode collapse and instability, drastically reducing their utility.