Training Language GANs from Scratch

      Cyprien de Masson d’Autume Mihaela Rosca Jack Rae Shakir Mohamed
                    DeepMind
            {cyprien,mihaelacr,jwrae,shakir}@google.com

                        Abstract
       Generative Adversarial Networks (GANs) enjoy great success at image genera-
       tion, but have proven difficult to train in the domain of natural language. Chal-
       lenges with gradient estimation, optimization instability, and mode collapse have
       lea ...