PDD: Partitioning DAG-Topology DNNs for Streaming Tasks

doi:10.1109/JIOT.2023.3323520

	PDD: Partitioning DAG-Topology DNNs for Streaming Tasks
	Liantao Wu 1; Guoliang Gao2 ; Jing Yu2 ; Fangtong Zhou2 ; Yang Yang 3; Tengfei Wang 4
	2023
发表期刊	IEEE INTERNET OF THINGS JOURNAL (IF:8.2[JCR-2023],9.0[5-Year])
ISSN	2372-2541
EISSN	2327-4662
卷号	PP 期号:99 页码:1-1
发表状态	已发表
DOI	10.1109/JIOT.2023.3323520
摘要	To enable the inference of high-precision deep neural networks (DNNs) on resource-constrained devices, DNN offloading has been widely explored in recent years. Some works have also integrated the chain-topology DNN offloading with pipeline processing to further reduce inference delay when processing streaming tasks. To improve the accuracy of the inference results, the topology of DNN tends to evolve from chain topology to directed acyclic graph (DAG) topology. However, most of the existing works do not study partitioning and offloading DAG-topology DNNs (DDNNs) for streaming tasks. Moreover, when partitioning computationally expensive DNN models, multi-partitioning probably outperforms the bi-partitioning method, and most of the works do not study multi-partitioning DAG-topology DNNs. In this paper, we propose a more general multi-partitioning and offloading method for large-scale DDNNs to process streaming tasks, which can adaptively partition DDNNs into multiple parts considering the computing power and bandwidth of all available computing units. Specifically, we first present a transforming method based on topological sorting that can losslessly transform DAG-topology DNNs into chain-topology DNNs (CDNNs). Then, based on greedy and dichotomy ideas, a multi-partitioning algorithm is designed to partition and offload CDNNs. In this way, we can solve DDNNs’ multi-partitioning problem based on the proposed transforming and partitioning algorithms. Experiments show that the method proposed in this paper significantly outperforms bi-partitioning and non-partitioning methods when offloading computationally expensive DNN models. IEEE
关键词	DNN partitioning streaming tasks DAG topology
URL	查看原文
收录类别	EI
语种	英语
出版者	Institute of Electrical and Electronics Engineers Inc.
EI入藏号	20234314970933
EI主题词	Directed graphs
EI分类号	461.4 Ergonomics and Human Factors Engineering ; 544.1 Copper ; 722.2 Computer Peripheral Equipment ; 722.4 Digital Computers and Systems ; 723 Computer Software, Data Handling and Applications ; 921.4 Combinatorial Mathematics, Includes Graph Theory, Set Theory
原始文献类型	Article in Press
来源库	IEEE
引用统计	正在获取...
文献类型	期刊论文
条目标识符	https://kms.shanghaitech.edu.cn/handle/2MSLDSTB/340943
专题	信息科学与技术学院信息科学与技术学院_硕士生
作者单位	1.the Shanghai Key Laboratory of Trustworthy Computing, Software Engineering Institute, East China Normal University, Shanghai, China 2.the Shanghai Institute of Microsystem and Information Technology, School of Information Science and Technology, ShanghaiTech University, Chinese Academy of Sciences, and University of Chinese Academy of Sciences, Shanghai, China 3.IoT Thrust at the Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China 4.State Key Laboratory of Maritime Technology and Safety, and School of Transportation and Logistics Engineering, Wuhan University of Technology, Wuhan, China
推荐引用方式 GB/T 7714	Liantao Wu,Guoliang Gao,Jing Yu,et al. PDD: Partitioning DAG-Topology DNNs for Streaming Tasks[J]. IEEE INTERNET OF THINGS JOURNAL,2023,PP(99):1-1.
APA	Liantao Wu,Guoliang Gao,Jing Yu,Fangtong Zhou,Yang Yang,&Tengfei Wang.(2023).PDD: Partitioning DAG-Topology DNNs for Streaming Tasks.IEEE INTERNET OF THINGS JOURNAL,PP(99),1-1.
MLA	Liantao Wu,et al."PDD: Partitioning DAG-Topology DNNs for Streaming Tasks".IEEE INTERNET OF THINGS JOURNAL PP.99(2023):1-1.