Quantized CNN: A Unified Approach to Accelerate and Compress Convolutional Networks

doi:10.1109/TNNLS.2017.2774288

	Quantized CNN: A Unified Approach to Accelerate and Compress Convolutional Networks
	Cheng, Jian3,4,5 ; Wu, Jiaxiang2,3,4 ; Leng, Cong3,4 ; Wang, Yuhang1,3,4 ; Hu, Qinghao3,4
刊名	IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
	2018-10-01
卷号	29 期号:10 页码:4730-4743
关键词	Acceleration And Compression Convolutional Neural Network (Cnn) Mobile Devices Product Quantization
ISSN号	2162-237X
DOI	10.1109/TNNLS.2017.2774288
文献子类	Article
英文摘要	We are witnessing an explosive development and widespread application of deep neural networks (DNNs) in various fields. However, DNN models, especially a convolutional neural network (CNN), usually involve massive parameters and are computationally expensive, making them extremely dependent on high-performance hardware. This prohibits their further extensions, e.g., applications on mobile devices. In this paper, we present a quantized CNN, a unified approach to accelerate and compress convolutional networks. Guided by minimizing the approximation error of individual layer's response, both fully connected and convolutional layers are carefully quantized. The inference computation can be effectively carried out on the quantized network, with much lower memory and storage consumption. Quantitative evaluation on two publicly available benchmarks demonstrates the promising performance of our approach: with comparable classification accuracy, it achieves 4 to 6x acceleration and 15 to 20x compression. With our method, accurate image classification can even be directly carried out on mobile devices within 1 s.
WOS关键词	LEARNING BINARY-CODES ; ITERATIVE QUANTIZATION ; PROCRUSTEAN APPROACH ; RECOGNITION ; FPGAS
WOS研究方向	Computer Science ; Engineering
语种	英语
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
WOS记录号	WOS:000445351300015
资助机构	National Natural Science Foundation of China(61332016) ; Scientific Research Key Program of Beijing Municipal Commission of Education(KZ201610005012) ; Fund of Hubei Key Laboratory of Transportation Internet of Things ; Fund of Jiangsu Key Laboratory of Big Data Analysis Technology
内容类型	期刊论文
源URL	[http://ir.ia.ac.cn/handle/173211/27920]
专题	中国科学院自动化研究所
通讯作者	Cheng, Jian
作者单位	1.UISEE Technol Beijing Ltd, Beijing 102402, Peoples R China 2.Tencent AI Lab, Machine Learning Grp, Shenzhen 518000, Peoples R China 3.Chinese Acad Sci, Inst Automat, Beijing 100190, Peoples R China 4.Univ Chinese Acad Sci, Beijing 100190, Peoples R China 5.CAS Ctr Excellence Brain Sci & Intelligence Techn, Beijing 100190, Peoples R China
推荐引用方式 GB/T 7714	Cheng, Jian,Wu, Jiaxiang,Leng, Cong,et al. Quantized CNN: A Unified Approach to Accelerate and Compress Convolutional Networks[J]. IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS,2018,29(10):4730-4743.
APA	Cheng, Jian,Wu, Jiaxiang,Leng, Cong,Wang, Yuhang,&Hu, Qinghao.(2018).Quantized CNN: A Unified Approach to Accelerate and Compress Convolutional Networks.IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS,29(10),4730-4743.
MLA	Cheng, Jian,et al."Quantized CNN: A Unified Approach to Accelerate and Compress Convolutional Networks".IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS 29.10(2018):4730-4743.