BriefGPT.xyz
May, 2019
从零开始训练语言GANs
Training language GANs from Scratch
HTML
PDF
Cyprien de Masson d'Autume, Mihaela Rosca, Jack Rae, Shakir Mohamed
TL;DR
本研究通过采用大批处理、密集奖励和鉴别器规则等现有技术来稳定和改善语言生成对抗网络,从而成功实现了ScratchGAN的从零训练,并表明其在语料库EMNLP2017 News和WikiText-103上的质量和多样性指标相当于最大似然训练。
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
,
→