探究GPU对神经网络的加速作用  

The Exploration of Accelerating Effects on Convolutional Neural Networks by GPU

作　　者：朱永贵[1] 华敏杰 张佳佳 ZHU Yong-gui;HUA Min-jie;ZHANG Jia-jia(School of Science,Communication University of China,Beijing 100024,China)

机构地区：[1]中国传媒大学理学院,北京100024

出　　处：《中国传媒大学学报（自然科学版）》2018年第5期73-76,共4页Journal of Communication University of China：Science and Technology

基　　金：国家自然科学基金项目(NO.11571325):基于压缩感知的核磁共振成像问题驱动的应用数学研究

摘　　要：随着人工智能的不断发展,神经网络在计算机视觉和自然语言处理中的应用越来越广泛。但是,由于神经网络的计算密集,单用CPU进行运算非常耗时,因此,通过GPU对神经网络进行加速具有很强的现实意义。本文首先对GPU的发展现状、物理架构及其与CPU在计算特性上的差别进行阐述,然后探究其对神经网络的加速作用。With the continuous development of artificial intelligence,neural networks have become more and more widely used in computer vision and natural language processing. However,due to the intensive computation of the neural network,it is very time-consuming to use only the CPU to perform operations. Therefore,it is of great practical significance to accelerate the neural network through the GPU. This article first elaborates on the status of GPU development,the physical architecture,and the differences with CPU in computing characteristics,and then explores its accelerating effects on neural networks.

关 键 词：GPU 神经网络 并行计算 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]