BriefGPT.xyz
Apr, 2014
用一种奇怪的技巧并行卷积神经网络
One weird trick for parallelizing convolutional neural networks
HTML
PDF
Alex Krizhevsky
TL;DR
介绍了一种新的方法，可在多个GPU上并行训练卷积神经网络，该方法在应用于现代卷积神经网络时比所有其他方法的扩展性都更好。
Abstract
I present a new way to
parallelize
the
training
of
convolutional neural networks
across
→