Adversarial network for natural language synthesis

Description

The key issue with generative tasks is about deciding what a good cost function should be. GAN introduces two networks to solve that. The generator network creates fake samples, and the discriminator network distinguishes them from real samples. GAN has been predominantly applied in image augmentation and is particularly good at generating continuous samples, but because of this, it can’t be used directly for text generation (as it’s sequence of discrete numbers.)

Rajib Biswas outlines the recent breakthroughs in applying adversarial networks for language generation such as SeqGAN (policy gradient reinforcement learning methods), LeakGAN (long text generation with leaked information), and a reparameterization trick for latent variables. You’ll learn about a variety of applications and tasks, including GAN for machine translation, GAN for dialogue generation, and GAN for style transfer, along with seeing a demonstration of language generation application with code.