NEW GAN BASELINE

Summary from the TLDR AI newsletter

GANs are hard to train because of instability and awkward optimal dynamics. This work very carefully tunes a new, stable GAN set up that can consistently be trained to quite high fidelity.