BriefGPT.xyz
May, 2019
基于乐观和锚定的随机梯度方法在极小极大问题上的ODE分析
ODE Analysis of Stochastic Gradient Methods with Optimism and Anchoring for Minimax Problems and GANs
HTML
PDF
Ernest K. Ryu, Kun Yuan, Wotao Yin
TL;DR
本文研究生成式对抗网络的训练动态及其变种中的梯度下降算法的极小极大博弈，通过微分方程的连续时间分析，研究了凸、凹假设下的最后迭代收敛性，并提出了具有悲观特征和锚定特征的simGD算法和新的anchored simGD算法。
Abstract
Despite remarkable empirical success, the
training dynamics
of
generative adversarial networks
(GAN), which involves solving a minimax game using stochastic gradients, is still poorly understood. In this work, we
→