卷积神经网络在GPU上的实现
首发时间:2015-11-10
摘要:卷积神经网络是目前机器学习领域的热点,其在图像识别领域有着广泛的应用。虽然卷积神经网络在各个方面都有不错的表现,但实际运用时不可避免会遇到卷积神经网络运算量过大,运算效率过低的问题。近几年GPU通用并行计算技术得到了飞速的发展,因为其计算核心的数量远远多于CPU,在高度并行的计算问题上,GPU的计算效率要远高于CPU。而最近英伟达公司在其GPU通用计算架构CUDA上推出了深度学习库CUDNN,该库为深度学习中常用的方法提供了接口,使得深度学习算法在GPU上的简易开发成为可能。本文借助CUDNN库,在GPU上实现了卷积神经网络,并测试了其相对于CPU的加速效果。
For information in English, please click here
The realization of the convolutional neural network on GPU
Abstract:Convolutional neural network is a hot spot in the field of machine learning. It has been widely used in the field of image recognition. Although convolution neural network has good performance, but when face practical problems, sometimes the calculation scale of the convolutional neural network is too large and computation efficiency is low. In recent years, the general parallel computing technology of GPU has been developed rapidly, because the comput core number of GPU is much more than that of CPU. Therefore, when facing with highly parallel computing problems, the capability of GPU is far beyond the CPU. Recently NVIDIA launch its deep learning library CUDNN base on its general computing architecture of GPU. The library provides interfaces for common operations of deep learning, making it simple to develop deep learning applications on GPU. This paper use CUDNN library to realize the convolutional neural network on GPU, and test the acceleration effect relative to CPU.
Keywords: Convolutional neural network GPU acceleration CUDNN
基金:
论文图表:
引用
No.4659458110179214****
同行评议
共计0人参与
勘误表
卷积神经网络在GPU上的实现
评论
全部评论0/1000