DyPipe: A Holistic Approach to Accelerating Dynamic Neural Networks with Dynamic Pipelining

doi:10.1007/s11390-021-1161-y

	DyPipe: A Holistic Approach to Accelerating Dynamic Neural Networks with Dynamic Pipelining
	Zhuang, Yi-Min 1,2; Hu, Xing 1; Chen, Xiao-Bing 1,2; Zhi, Tian 1
	2023-07-01
发表期刊	JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY
ISSN	1000-9000
卷号	38 期号:4 页码:899-910
摘要	Dynamic neural network (NN) techniques are increasingly important because they facilitate deep learning techniques with more complex network architectures. However, existing studies, which predominantly optimize the static computational graphs by static scheduling methods, usually focus on optimizing static neural networks in deep neural network (DNN) accelerators. We analyze the execution process of dynamic neural networks and observe that dynamic features introduce challenges for efficient scheduling and pipelining in existing DNN accelerators. We propose DyPipe, a holistic approach to optimizing dynamic neural network inferences in enhanced DNN accelerators. DyPipe achieves significant performance improvements for dynamic neural networks while it introduces negligible overhead for static neural networks. Our evaluation demonstrates that DyPipe achieves 1.7x speedup on dynamic neural networks and maintains more than 96% performance for static neural networks.
关键词	dynamic neural network (NN) deep neural network (DNN) accelerator dynamic pipelining
DOI	10.1007/s11390-021-1161-y
收录类别	SCI
语种	英语
资助项目	Beijing Natural Science Foundation[JQ18013] ; National Natural Science Foundation of China[61925208] ; National Natural Science Foundation of China[61732007] ; National Natural Science Foundation of China[61732002] ; National Natural Science Foundation of China[61906179] ; Strategic Priority Research Program of Chinese Academy of Sciences (CAS)[XDB32050200] ; Youth Innovation Promotion Association CAS ; Beijing Academy of Artificial Intelligence (BAAI) ; Xplore Prize
WOS研究方向	Computer Science
WOS类目	Computer Science, Hardware & Architecture ; Computer Science, Software Engineering
WOS记录号	WOS:001102032000012
出版者	SPRINGER SINGAPORE PTE LTD
引用统计	被引频次：1[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/2XEOYT63/38066
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Zhi, Tian
作者单位	1.Chinese Acad Sci, Inst Comp Technol, State Key Lab Processors, Beijing 100190, Peoples R China 2.Univ Chinese Acad Sci, Beijing 100049, Peoples R China
推荐引用方式 GB/T 7714	Zhuang, Yi-Min,Hu, Xing,Chen, Xiao-Bing,et al. DyPipe: A Holistic Approach to Accelerating Dynamic Neural Networks with Dynamic Pipelining[J]. JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY,2023,38(4):899-910.
APA	Zhuang, Yi-Min,Hu, Xing,Chen, Xiao-Bing,&Zhi, Tian.(2023).DyPipe: A Holistic Approach to Accelerating Dynamic Neural Networks with Dynamic Pipelining.JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY,38(4),899-910.
MLA	Zhuang, Yi-Min,et al."DyPipe: A Holistic Approach to Accelerating Dynamic Neural Networks with Dynamic Pipelining".JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY 38.4(2023):899-910.