BriefGPT.xyz
Oct, 2016
卷积神经网络中的子流形卷积核优化
Optimization on Submanifolds of Convolution Kernels in CNNs
HTML
PDF
Mete Ozay, Takayuki Okatani
TL;DR
本文提出了新的核规范化方法,解释了该方法对CNN中核搜索空间的几何形状的影响,并证明了该方法几乎可以保证收敛于CNN分类损失的单一最小值,为图像分类基准测试提供了最先进的性能。
Abstract
kernel normalization
methods have been employed to improve robustness of optimization methods to reparametrization of convolution kernels, covariate shift, and to accelerate training of
convolutional neural networks
→