BriefGPT.xyz
May, 2017
使用多个随机投影稳定 GAN 训练
Stabilizing GAN Training with Multiple Random Projections
HTML
PDF
Behnam Neyshabur, Srinadh Bhojanapalli, Ayan Chakrabarti
TL;DR
本文提出了一种针对高维度情况下 GAN 训练不稳定的方法,即使用一个生成器同时生成多个不同低维度投影数据的判别器来提供连续的梯度,实验结果表明该方法能够生成较高质量的图像样本。
Abstract
Training
generative adversarial networks
is unstable in
high-dimensions
when the true data distribution lies on a lower-dimensional manifold. The
→