ABCP: Automatic Blockwise and Channelwise Network Pruning via Joint Search

doi:10.1109/TCDS.2022.3230858

CORC > 自动化研究所 > 中国科学院自动化研究所 > 复杂系统管理与控制国家重点实验室 > 深度强化学习

	ABCP: Automatic Blockwise and Channelwise Network Pruning via Joint Search
	Li, Jiaqi 2,3; Li, Haoran 1,3; Chen, Yaran 1,3; Ding, Zixiang 1,3; Li, Nannan 1,3; Ma, Mingjun 1,3; Duan, Zicheng ; Zhao, Dongbin 1,3
刊名	IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS
	2023-09-01
卷号	15 期号:3 页码:1560-1573
关键词	Joint search model compression pruning reinforcement learning
ISSN号	2379-8920
DOI	10.1109/TCDS.2022.3230858
通讯作者	Li, Haoran(lihaoran2015@ia.ac.cn) ; Chen, Yaran(chenyaran2013@ia.ac.cn)
英文摘要	Currently, an increasing number of model pruning methods are proposed to resolve the contradictions between the computer powers required by the deep learning models and the resource-constrained devices. However, for simple tasks like robotic detection, most of the traditional rule-based network pruning methods cannot reach a sufficient compression ratio with low accuracy loss and are time consuming as well as laborious. In this article, we propose automatic blockwise and channel-wise network pruning (ABCP) to jointly search the blockwise and channelwise pruning action for robotic detection by deep reinforcement learning. A joint sample algorithm is proposed to simultaneously generate the pruning choice of each residual block and the channel pruning ratio of each convolutional layer from the discrete and continuous search space, respectively. The best pruning action taking both the accuracy and the complexity of the model into account is obtained finally. Compared with the traditional rule-based pruning method, this pipeline saves human labor and achieves a higher compression ratio with lower accuracy loss. Tested on the mobile robot detection data set, the pruned YOLOv3 model saves 99.5% floating-point operations, reduces 99.5% parameters, and achieves 37.3x speed up with only 2.8% mean of average precision (mAP) loss. On the sim2real detection data set for robotic detection task, the pruned YOLOv3 model achieves 9.6% better mAP than the baseline model, showing better robustness performance.
资助项目	National Natural Science Foundation of China (NSFC)[62006226] ; National Natural Science Foundation of China (NSFC)[62103409] ; Strategic Priority Research Program of Chinese Academy of Sciences[XDA27030400]
WOS关键词	ATTENTION
WOS研究方向	Computer Science ; Robotics ; Neurosciences & Neurology
语种	英语
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
WOS记录号	WOS:001089186500047
资助机构	National Natural Science Foundation of China (NSFC) ; Strategic Priority Research Program of Chinese Academy of Sciences
内容类型	期刊论文
源URL	[http://ir.ia.ac.cn/handle/173211/54301]
专题	复杂系统管理与控制国家重点实验室_深度强化学习
通讯作者	Li, Haoran; Chen, Yaran
作者单位	1.Univ Chinese Acad Sci, Sch Artificial Intelligence, Beijing 101408, Peoples R China 2.Beijing Inst Technol, Sch Mech Engn, Beijing 100081, Peoples R China 3.Chinese Acad Sci, Inst Automat, State Key Lab Management & Control Complex Syst, Beijing 100190, Peoples R China
推荐引用方式 GB/T 7714	Li, Jiaqi,Li, Haoran,Chen, Yaran,et al. ABCP: Automatic Blockwise and Channelwise Network Pruning via Joint Search[J]. IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS,2023,15(3):1560-1573.
APA	Li, Jiaqi.,Li, Haoran.,Chen, Yaran.,Ding, Zixiang.,Li, Nannan.,...&Zhao, Dongbin.(2023).ABCP: Automatic Blockwise and Channelwise Network Pruning via Joint Search.IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS,15(3),1560-1573.
MLA	Li, Jiaqi,et al."ABCP: Automatic Blockwise and Channelwise Network Pruning via Joint Search".IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS 15.3(2023):1560-1573.