In our lab meeting this week, Nishant will talk about some clever adversarial training tricks for sequence generation tasks. The title and abstract of his talk:
Abstract: Adversarial methods of training typically employ a “discriminator” that criticizes some aspect of the predicted output of a generative model. Although discriminators are a key component in the seemingly complex generative adversarial network (GAN) or adversial feature learning frameworks, they can be used in quite simple but clever ways to control sequence generation (especially). In this discussion, we will see two such applications to language modelling and machine translation.
Relevant Papers:
Wednesday, January 16th, 11:00 a.m. TASC1 9408.
