Toward Efficient Execution of Mainstream Deep Learning Frameworks on Mobile Devices: Architectural Implications

doi:10.1109/TCAD.2020.3003233

	Toward Efficient Execution of Mainstream Deep Learning Frameworks on Mobile Devices: Architectural Implications
	Dai, Yuting 1,2,3; Zhang, Rui 2,4; Xue, Rui 2,5; Liu, Benyong 3; Li, Tao 2
	2021-03-01
发表期刊	IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS
ISSN	0278-0070
卷号	40 期号:3 页码:453-466
摘要	In recent years, continuous growing interests have been seen in bringing artificial intelligence capabilities to mobile devices. However, the related work still faces several issues, such as constrained computation and memory resources, power drain, and thermal limitation. To develop deep learning (DL) algorithms on mobile devices, we need to understand their behaviors. In this article, we explore the architectural behaviors of some mainstream DL frameworks on mobile devices by performing a comprehensive characterization of performance, accuracy, energy efficiency, and thermal behaviors. We experimentally choose four model compression methods to perform on networks and in addition, analyze the related impact on the nodes amount, memory, execution time, model size, inference time, energy consumption, and thermal distribution. With insights into DL-based mobile application characteristics, we hope to guide the design of future smartphone platforms for lower energy consumption.
关键词	Deep learning Mobile handsets Performance evaluation Mobile applications Computational modeling Quantization (signal) Central Processing Unit CPU usage deep learning (DL) energy efficiency microarchitecture mobile phone network compression thermal characterization
DOI	10.1109/TCAD.2020.3003233
收录类别	SCI
语种	英语
WOS研究方向	Computer Science ; Engineering
WOS类目	Computer Science, Hardware & Architecture ; Computer Science, Interdisciplinary Applications ; Engineering, Electrical & Electronic
WOS记录号	WOS:000621402700005
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
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/2XEOYT63/16904
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Dai, Yuting
作者单位	1.Chengdu Normal Univ, Sch Comp Sci, Chengdu 610041, Peoples R China 2.Univ Florida, Dept ECE, Gainesville, FL 32611 USA 3.Guizhou Univ, Coll Comp Sci & Technol, Guiyang 550025, Peoples R China 4.Nankai Univ, Coll Comp Sci, Tianjin 300071, Peoples R China 5.Chinese Acad Sci, Inst Comp Technol, State Key Lab Comp Architecture, Beijing 100190, Peoples R China
推荐引用方式 GB/T 7714	Dai, Yuting,Zhang, Rui,Xue, Rui,et al. Toward Efficient Execution of Mainstream Deep Learning Frameworks on Mobile Devices: Architectural Implications[J]. IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS,2021,40(3):453-466.
APA	Dai, Yuting,Zhang, Rui,Xue, Rui,Liu, Benyong,&Li, Tao.(2021).Toward Efficient Execution of Mainstream Deep Learning Frameworks on Mobile Devices: Architectural Implications.IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS,40(3),453-466.
MLA	Dai, Yuting,et al."Toward Efficient Execution of Mainstream Deep Learning Frameworks on Mobile Devices: Architectural Implications".IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS 40.3(2021):453-466.