CSCC: Convolution Split Compression Calculation Algorithm for Deep Neural Network

doi:10.1109/ACCESS.2019.2919554

	CSCC: Convolution Split Compression Calculation Algorithm for Deep Neural Network
	Fan, Shengyu 1; Yu, Hui 2; Lu, Dianjie 1; Jiao, Shuai 3; Xu, Weizhi 1; Liu, Fangai 1; Liu, Zhiyong 3
	2019
发表期刊	IEEE ACCESS
ISSN	2169-3536
卷号	7 页码:71607-71615
摘要	Convolutional Neural Networks (CNNs) have become one of the most successful machine learning techniques for image and video processing. The most computationally intensive part of the CNN is the convolutional layers, which have the multi-channel image and multiple kernels. However, due to the network pruning operation and the application of RELU activation function operation in the training process, numerous zero values are generated in the network. This paper proposes the convolution split compression calculation (CSCC) algorithm, which improves the performance of the convolution layer by utilizing the sparse characteristic of the feature map. In the CSCC algorithm, first, the feature map is directly converted into a sparse matrix of compressed sparse row (CSR) format, which avoids expanding feature map to an intermediate matrix and reduces the memory space consumption. Second, the convolution kernel is converted into a vector. Finally, the convolution result is obtained by the sparse matrix vector multiplication (SpMV). The experimental results show that the CSCC algorithm has a good advantage in computation speed and memory consumption compared with the other convolution algorithms.
关键词	Convolutional neural network sparse matrix vector multiplication neural networks convolution sparse matrices
DOI	10.1109/ACCESS.2019.2919554
收录类别	SCI
语种	英语
资助项目	NNSF of China[61602285] ; NNSF of China[61602284] ; NNSF of China[61772321] ; NNSF of China[61520106005] ; National Key Research and Development Program of China[2017YFB1010000] ; Shandong Natural Science Foundation[ZR2015FQ009] ; Shandong Natural Science Foundation[ZR2016FP07] ; Shandong Natural Science Foundation[ZR2019MF017] ; Project of Shandong Province Higher Educational Science and Technology Program[J16LN05]
WOS研究方向	Computer Science ; Engineering ; Telecommunications
WOS类目	Computer Science, Information Systems ; Engineering, Electrical & Electronic ; Telecommunications
WOS记录号	WOS:000471033300001
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
引用统计	被引频次：8[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/4220
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Yu, Hui; Xu, Weizhi
作者单位	1.Shandong Normal Univ, Sch Informat Sci & Engn, Jinan 250358, Shandong, Peoples R China 2.Shandong Normal Univ, Sch Business, Jinan 250358, Shandong, Peoples R China 3.Chinese Acad Sci, Inst Comp Technol, Beijing 100190, Peoples R China
推荐引用方式 GB/T 7714	Fan, Shengyu,Yu, Hui,Lu, Dianjie,et al. CSCC: Convolution Split Compression Calculation Algorithm for Deep Neural Network[J]. IEEE ACCESS,2019,7:71607-71615.
APA	Fan, Shengyu.,Yu, Hui.,Lu, Dianjie.,Jiao, Shuai.,Xu, Weizhi.,...&Liu, Zhiyong.(2019).CSCC: Convolution Split Compression Calculation Algorithm for Deep Neural Network.IEEE ACCESS,7,71607-71615.
MLA	Fan, Shengyu,et al."CSCC: Convolution Split Compression Calculation Algorithm for Deep Neural Network".IEEE ACCESS 7(2019):71607-71615.