How (not) to Train your Generative Model: Scheduled Sampling, Likelihood, Adversary?

Open in new window