Training language GANs from Scratch

TL;DR

This paper demonstrates that it is feasible to train language GANs from scratch without pre-training, using techniques like large batch sizes and regularization, achieving comparable results to traditional methods.

Contribution

The authors introduce ScratchGAN, a method for training language GANs from scratch, overcoming previous challenges and eliminating the need for pre-training.

Findings

ScratchGAN achieves comparable quality to maximum likelihood models.

Techniques like large batch sizes and discriminator regularization improve stability.

The approach works on EMNLP2017 News and WikiText-103 datasets.

Abstract

Generative Adversarial Networks (GANs) enjoy great success at image generation, but have proven difficult to train in the domain of natural language. Challenges with gradient estimation, optimization instability, and mode collapse have lead practitioners to resort to maximum likelihood pre-training, followed by small amounts of adversarial fine-tuning. The benefits of GAN fine-tuning for language generation are unclear, as the resulting models produce comparable or worse samples than traditional language models. We show it is in fact possible to train a language GAN from scratch -- without maximum likelihood pre-training. We combine existing techniques such as large batch sizes, dense rewards and discriminator regularization to stabilize and improve language GANs. The resulting model, ScratchGAN, performs comparably to maximum likelihood training on EMNLP2017 News and WikiText-103 corpora according to quality and diversity metrics.