BriefGPT.xyz
Nov, 2016
多元神经网络学习真实目标函数
Diversity Leads to Generalization in Neural Networks
HTML
PDF
Bo Xie, Yingyu Liang, Le Song
TL;DR
通过对具有ReLU激活函数的一层神经网络的分析,我们发现神经网络具有良好的优化特性,其具有多样的单元没有虚假局部最小值,在满足“扩展特征矩阵”的最小奇异值足够大的条件下,可以使损失函数变得任意小。
Abstract
neural networks
are a powerful class of functions that can be trained with simple
gradient descent
to achieve state-of-the-art performance on a variety of applications. Despite their practical success, there is a
→