BriefGPT.xyz
Dec, 2019
LOGAN: 生成对抗网络的潜在优化
LOGAN: Latent Optimisation for Generative Adversarial Networks
HTML
PDF
Yan Wu, Jeff Donahue, David Balduzzi, Karen Simonyan, Timothy Lillicrap
TL;DR
利用自然梯度基于潜在空间的优化改进了CS-GAN模型，增强了模型的生成器和辨别器之间的互动，通过在ImageNet数据集上的实验表明，该方法在生成对抗网络训练中可显著提高模型性能。
Abstract
Training
generative adversarial networks
requires balancing of delicate
adversarial dynamics
. Even with careful tuning, training may diverge or end up in a bad equilibrium with dropped modes. In this work, we int
→