BriefGPT.xyz
Aug, 2018
用傅里叶分析理解深度学习中的训练和泛化
Understanding training and generalization in deep learning by Fourier analysis
HTML
PDF
Zhiqin John Xu
TL;DR
通过傅里叶分析，研究DNN训练的理论框架，解释了梯度下降法训练DNN经常赋予目标函数低频分量更高的优先级，小的初始化导致DNN具有良好的泛化能力，同时保留拟合任何函数的能力。这些结果进一步得到了DNN拟合自然图像、一维函数和MNIST数据集的实验证实。
Abstract
Background: It is still an open research area to theoretically understand why
deep neural networks
(DNNs)---equipped with many more parameters than training data and trained by (stochastic)
gradient-based methods
→