摘要
近年来,深度卷积神经网络在图像识别和语音识别等领域被广泛运用,取得了很好的效果。深度卷积神经网络是层数较多的卷积神经网络,有数千万参数需要学习,计算开销大,导致训练非常耗时。针对这种情况,本文提出深度卷积神经网络的多GPU并行框架,设计并实现模型并行引擎,依托多GPU的强大协同并行计算能力,结合深度卷积神经网络在训练中的并行特点,实现快速高效的深度卷积神经网络训练。
In recent years, deep convolutional neural network is widely used in the fields of image recognition and speech recogni- tion, and achieves good results. Deep convolutional neural networks are the convolutional neural networks with multiple layers, tens of millions of parameters need to be learned, and computational overhead is large, so the training is very time-consuming. In view of this situation, we propose a muhi-GPU parallel framework of deep convolutional neural networks, design and implement model parallel engine, relying on the powerful collaborative parallel computing ability of muhi-GPU, combined with the parallel characteristics of deep convolutional neural networks in training, to achieve fast and efficient deep convolution neural networks training.
出处
《计算机与现代化》
2016年第11期95-98,共4页
Computer and Modernization
基金
国家自然科学基金青年基金资助项目(61202136)
关键词
深度卷积神经网络
GPU
并行框架
图像识别
大数据
deep convolutional neural networks
graphic processing unit
parallel framework
image recognition
big data