A Decomposable Winograd Method for N-D Convolution Acceleration in Video Analysis

doi:10.1007/s11263-021-01500-9

	A Decomposable Winograd Method for N-D Convolution Acceleration in Video Analysis
	Huang, Di 1,3,4; Zhang, Rui 1,4; Zhang, Xishan 1,4; Wu, Fan 2,3,4; Wang, Xianzhuo 1,3,4; Jin, Pengwei 1,3,4; Liu, Shaoli 4; Li, Ling 2; Chen, Yunji 1,3
	2021-08-04
发表期刊	INTERNATIONAL JOURNAL OF COMPUTER VISION
ISSN	0920-5691
页码	21
摘要	Winograd's minimal filtering algorithm has been widely used in 2-D Convolutional Neural Networks (CNNs) to reduce the number of multiplications for faster processing. However, it is only effective on convolutions with kernel size as 3 and stride as 1, because it suffers from significantly increased FLOPs and numerical accuracy problems for kernel size larger than 3 and fails on convolution with stride larger than 1. Worse, the extension to N-D convolution will intensify the numerical accuracy problem. These problems severely obstruct Winograd's minimal filtering algorithm's application to video analysis. In this paper, we propose a novel Decomposable Winograd Method (DWM) for the N-D convolution acceleration, which breaks through the limitation of original Winograd's minimal filtering algorithm to more general convolutions. DWM decomposes kernels with large size or stride>1 to several small kernels with stride as 1 for further applying Winograd algorithm, so that DWMcan reduce the number of multiplications while keeping the numerical accuracy. It enables the fast exploration of larger kernel size, larger stride value, and higher dimensions in CNNs for high performance and accuracy and even the potential for new CNNs. Comparing against the original Winograd algorithm, the proposed DWM is able to support all kinds of N-D convolutions with a speedup of 1.44x-3.38x, without affecting the numerical accuracy.
关键词	Convolution neural networks Model acceleration Winograd algorithm Video analysis
DOI	10.1007/s11263-021-01500-9
收录类别	SCI
语种	英语
资助项目	Beijing Natural Science Foundation[JQ18013] ; NSF of China[61925208] ; NSF of China[61906179] ; NSF of China[62002338] ; NSF of China[61732007] ; NSF of China[61732002] ; NSF of China[U19B2019] ; NSF of China[U20A20227] ; Strategic Priority Research Program of Chinese Academy of Science[XDB 32050200] ; Strategic Priority Research Program of Chinese Academy of Science[XDC05010300] ; Beijing Academy of Artificial Intelligence (BAAI) ; Beijing Nova Program of Science and Technology[Z191100001119093] ; Youth Innovation Promotion Association CAS
WOS研究方向	Computer Science
WOS类目	Computer Science, Artificial Intelligence
WOS记录号	WOS:000681171700002
出版者	SPRINGER
引用统计	被引频次：2[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/2XEOYT63/17395
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Zhang, Rui
作者单位	1.Chinese Acad Sci, Inst Comp Technol, SKL Comp Architecture, Beijing, Peoples R China 2.Chinese Acad Sci, Inst Software, Beijing, Peoples R China 3.Univ Chinese Acad Sci, Beijing, Peoples R China 4.Cambricon Tech Ltd, Beijing, Peoples R China
推荐引用方式 GB/T 7714	Huang, Di,Zhang, Rui,Zhang, Xishan,et al. A Decomposable Winograd Method for N-D Convolution Acceleration in Video Analysis[J]. INTERNATIONAL JOURNAL OF COMPUTER VISION,2021:21.
APA	Huang, Di.,Zhang, Rui.,Zhang, Xishan.,Wu, Fan.,Wang, Xianzhuo.,...&Chen, Yunji.(2021).A Decomposable Winograd Method for N-D Convolution Acceleration in Video Analysis.INTERNATIONAL JOURNAL OF COMPUTER VISION,21.
MLA	Huang, Di,et al."A Decomposable Winograd Method for N-D Convolution Acceleration in Video Analysis".INTERNATIONAL JOURNAL OF COMPUTER VISION (2021):21.