BriefGPT.xyz
Jun, 2020
使用概率比剪切和样本重新加权改进GAN训练
Improving GAN Training with Probability Ratio Clipping and Sample Reweighting
HTML
PDF
Yue Wu, Pan Zhou, Andrew Gordon Wilson, Eric P. Xing, Zhiting Hu
TL;DR
提出了一种基于变分的生成式对抗网络训练框架，通过概率裁剪和样本重新加权等技术，提升文本生成、文本风格转移和图像生成等多个任务的表现
Abstract
Despite success on a wide range of problems related to vision,
generative adversarial networks
(GANs) can suffer from inferior performance due to unstable training, especially for
text generation
. we propose a ne
→