二值卷积神经网络综述

丁文锐; 刘春蕾; 李越; 张宝昌

doi:10.7527/S1000-6893.2020.24504

航空学报 >

2021 , Vol. 42 >Issue 6: 24504 - 024504

DOI: https://doi.org/10.7527/S1000-6893.2020.24504

综述

二值卷积神经网络综述

丁文锐 ,
刘春蕾 ,
李越 ,
张宝昌

展开

1. 北京航空航天大学无人系统研究院, 北京 100083;
2. 北京航空航天大学电子信息工程学院, 北京 100083;
3. 北京航空航天大学自动化科学与电气工程学院, 北京 100083

收稿日期: 2020-07-07

修回日期: 2020-08-03

网络出版日期: 2020-09-24

基金资助

国家自然科学基金企业创新发展联合基金（U20B2042）；国家自然科学基金（62076019）；国家重点研发计划"新一代人工智能专项（2030）"（2020AAA0108200）

收起

Binary convolutional neural network: Review

DING Wenrui ,
LIU Chunlei ,
LI Yue ,
ZHANG Baochang

Expand

1. Unmanned System Research Institute, Beihang University, Beijing 100083, China;
2. School of Electronic and Information Engineering, Beihang University, Beijing 100083, China;
3. School of Automation Science and Electrical Engineering, Beihang University, Beijing 100083, China

Received date: 2020-07-07

Revised date: 2020-08-03

Online published: 2020-09-24

Supported by

National Natural Science Foundation of China (U20B2042); National Natural Science Foundation of China (62076019); Science and Technology Innovation 2030-Key Project of "New Generation Artificial Intelligence" (2020AAA0108200)

Fold

摘要

二值卷积神经网络（BNN）占用存储空间小、计算效率高，然而由于网络前向的二值量化与反向梯度的不匹配问题，使其与同结构的全精度深度卷积神经网络（CNN）之间存在较大的性能差距，影响了其在资源受限平台上的部署。至今，研究者已提出了一系列网络设计与训练方法来降低卷积神经网络在二值化过程中的性能损失，以推动二值卷积神经网络在嵌入式便携设备发展中的应用。因此，本文对二值卷积神经网络进行综述，主要从提高网络表达能力与充分挖掘网络训练潜力两大方面，给出了当前二值卷积神经网络的发展脉络与研究现状。具体而言，提高网络表达能力分为二值化量化方法设计、结构设计两方面，充分挖掘网络训练潜力分为损失函数设计与训练策略两方面。最后，对二值卷积神经网络在不同任务与硬件平台的实验情况进行了总结和技术分析，并展望了未来研究中可能面临的挑战。

关键词： 二值卷积神经网络; 全精度卷积神经网络; 二值化; 量化; 模型压缩; 轻量化; 深度学习

本文引用格式

丁文锐 , 刘春蕾 , 李越 , 张宝昌 . 二值卷积神经网络综述[J]. 航空学报, 2021 , 42(6) : 24504 -024504 . DOI: 10.7527/S1000-6893.2020.24504

Abstract

In recent years, Binary Convolutional Neural Networks (BNNs) have attracted much attention owing to their low storage and high computational efficiency. However, the mismatch between forward and backward quantization results in a huge performance gap between the BNN and the full-precision convolutional neural network, affecting the deployment of the BNN on resource-constrained platforms. Researchers have proposed a series of algorithms and training methods to reduce the performance gap during the binarization process, thereby promoting the application of BNNs to embedded portable devices. This paper makes a comprehensive review of BNNs, mainly from the perspectives of improving network representative capabilities and fully exploring the network training potential. Specifically, improving network representative capabilities includes the design of the binary quantization method and structure design, while fully exploring the network training potential involves loss function design and the training strategy. Finally, we discuss the performance of BNNs in different tasks and hardware platforms, and summarize the challenges in future research.

Key words： binary convolutional neural networks; full-precision convolutional neural networks; binarization; quantization; model compression; lightweight; deep learning

参考文献

[1] HE K M, ZHANG X Y, REN S Q, et al. Deep residual iearning for image recognition[C]//Proceeding of IEEE conference on computer vision and pattern recognition. Piscataway:IEEE Press, 2016:770-778.
[2] REN S Q, HE K M, GIRSHICK R, et al. Faster RCNN:Towards real-time object detection with region proposal networks[C]//In Advances in Neural Information Processing Systems.Piscataway:IEEE Press, 2015:91-99.
[3] CHEN L C, PAPANDREOU G, KOKKINOS I, et al. Deeplab:Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017,40(4):834-848.
[4] 纪荣嵘, 林绍辉, 晁飞, 等. 深度神经网络压缩与加速综述[J]. 计算机研究与发展, 2018, 55(9):1871-1888. JI R R, LIN S H, CHAO F, et al. Deep neural network compression and acceleration:A review[J]. Journal of Computer Research and Development, 2018, 55(9):1871-1888(in Chinese).
[5] 雷杰, 高鑫, 宋杰, 等. 深度网络模型压缩综述[J]. 软件学报, 2018, 29(2):251-266. LEI J, GAO X, SONG J, et al. Survey of deep neural network model compression[J]. Journal of Software, 2018, 29(2):251-266(in Chinese).
[6] 李江昀, 赵义凯, 薛卓尔, 等. 深度神经网络模型压缩综述[J]. 工程科学学报, 2019, 41(10):1229-1239. LI J Y, ZHAO Y K, XUE Z E, et al. A survey of model compression for deep neural networks[J]. Chinese Journal of Engineering, 2019, 41(10):1229-1239(in Chinese).
[7] 曹文龙, 芮建武, 李敏. 神经网络模型压缩方法综述[J]. 计算机应用研究, 2019, 36(3):649-656. CAO W L, RUI J W, LI M. Summary of neural network model compression methods[J]. Application Research of Computers, 2019, 36(3):649-656(in Chinese).
[8] 耿丽丽, 牛保宁. 深度神经网络模型压缩综述[J]. 计算机科学与探索, 2020, 14(9):1-16. GENG L L, NIU B N. Survey of deep neural networks model compression[J]. Journal of Frontiers of Computer Science & Technology, 2020, 14(9):1-16(in Chinese).
[9] 张弛, 田锦, 王永森, 等. 神经网络模型压缩方法综述[C]//中国计算机用户协会网络应用分会2018年第二十二届网络新技术与应用年会论文集, 2018. ZHANG C, TIAN J, WANG Y S, et al. Overview of neural network model compression methods[C]//Proceedings of the 22nd Annual Conference of New Network Technologies and Applications in 2018 of the Network Application Branch of China Computer Users Association, 2018(in Chinese).
[10] 蔡瑞初, 钟椿荣, 余洋, 等. 面向"边缘"应用的卷积神经网络量化与压缩方法[J]. 计算机应用, 2018, 38(9):2449-2454. CAI R C, ZHONG C R, YU Y, et al. CNN quantization and compression strategy for edge computing applications[J]. Journal of Computer Applications, 2018, 38(9):2449-2454(in Chinese).
[11] 袁庆祝. 基于CNN卷积神经网络的图像压缩技术[D]. 兰州:兰州大学, 2019. YUAN Q Z. Image compression technology based on CNN convolutional neural network[D]. Lanzhou:Lanzhou University, 2019(in Chinese).
[12] RASTEGARI M, ORDONEZ V, REDMON J, et al. XNOR-Net:ImageNet classification using binary convolutional neural networks[C]//In Proceedings of the European Conference on Computer Vision, 2016:525-542.
[13] LIU Z C, WU B Y, LUO W H, et al. Bi-Real Net:Enhancing the performance of 1-bit CNNs with improved representational capability and advanced training algorithm[C]//Proceedings of the European Conference on Computer Vision, 2018:722-737.
[14] HOU L, YAO Q M, KWOK J T. Loss-aware binarization of deep convolutional networks[C]//Proceedings of the International Conference on Learning Representations, 2017.
[15] LI Z, NI B, ZHANG W, et al. Performance guaranteed network acceleration via high-order residual quantization[C]//Proceedings of IEEE International Conference on Computer Vision. Piscataway:IEEE Press, 2017.
[16] GU J X, LI C, ZHANG B C, et al. Projection convolutional neural networks for 1-bit CNNs via discrete back propagation[C]//Proceeding of the Conference of Association for the Advance of Artificial Intelligence, 2019:8344-8351.
[17] COURBARIAUX M, HUBARA I, SOUDRY D, et al. Binarized neural networks:Training deep neural networks with weights and activations constrained to +1 or -1[DB/OL]. arXiv preprint:1602.02830, 2016.
[18] MARTINEZ B, YANG J, BULAT A, et al. Training binary neural networks with real-to-binary convolutions[C]//Proceedings of the International Conference on Learning Representations, 2020.
[19] DARABI S, BELBAHRI M, COURVARIAUX M, et al. BNN+:Improved binary network training[C]//NeurIPS Workshop on Energy Efficient Machine Learning and Cognitive Computing, 2019.
[20] LIU C L, DING W R, XIA X, et al. Circulant binary convolutional networks:Enhancing the performance of 1-bit DCNNs with circulant back propagation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2019:2691-2699.
[21] ZHUANG B, SHEN C, TAN M, et al. Structured binary neural networks for accurate image classification and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2019:413-422.
[22] KIM H, KIM K, KIM J, et al. BinaryDuo:Reducing gradient mismatch in binary activation network by coupling binary activations[C]//Proceedings of the International Conference on Learning Representations, 2020.
[23] DING R Z, CHIN T W, LIU Z Y, et al. Regularizing activation distribution for training binarized deep networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2019:11400-11409.
[24] LIU Z C, SHEN Z Q, SAVVIDES M, et al. Reactnet:Towards precise binary neural network with generalized activation functions[C]//Proceedings of the European Conference on Computer Vision, 2020.
[25] GU J X, ZHAO J H, JIANG X L, et al. Bayesian optimized 1-bit CNNs[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway:IEEE Press, 2019:4908-4916.
[26] ZHUANG B H, SHEN C H, TAN M K, et al. Towards effective low-bitwidth convolutional neural networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. Piscataway:IEEE Press, 2018:7920-7928.
[27] MISHRA A, MARR D. Apprentice:Using knowledge distillation techniques to improve low-precision network accuracy[C]//International Conference on Learning Representations. Piscataway:IEEE Press, 2018.
[28] LIU C L, DING W R, XIA X, et al. RBCN:Rectified binary convolutional networks for enhancing the performance of 1-bit DCNNs[C]//Proceeding of International Joint Conference on Artificial Intelligence, 2019.
[29] ZHOU S C, WU Y X, NI Z K, et al. Dorefa-net:Training low bitwidth convolutional neural networks with low bitwidth gradients[DB/OL].arXiv preprint:1606.06160, 2016.
[30] Nihui, BUG1989, Howave, gemfield, Corea, and eric612.ncnn[EB/OL].[2020-06-15]. https://github.com/Tencent/ncnn.
[31] QIN H T, GONG R H, LIU X L, et al. Forward and backward information retention for accurate binary neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2020:2250-2259.
[32] WANG Z W, WU Z Y, LU J W, et al. BiDet:An efficient binarized object detector[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2020.
[33] LIU H X, SIMONYAN K, YANG Y M. Darts:Differentiable architecture search[DB/OL]. arXiv preprint:1806.09055, 2018.
[34] CHEN H L, ZHUO L A, ZHANG B C, et al. Binarized neural architecture search[C]//Proceeding of the Conference of Association for the Advance of Artificial Intelligence, 2020.
[35] SHEN M Z, HAN K, XU C J, et al. Searching for accurate binary neural architectures[C]//Proceedings of the IEEE International Conference on Computer Vision Workshops. Piscataway:IEEE Press, 2019.
[36] HELWEGEN K, WIDDICOMBE J, GEIGER L, et al. Latent weights do not exist:Rethinking binarized neural network optimization[C]//Proceeding of the Conference of Advances in Neural Information Processing Systems, 2019:7531-7542.
[37] FRIESEN L A, DOMINGOS P. Deep learning as a mixed convex-combinatorial optimization problem[C]//Proceeding of the International Conference on Learning Representations, 2018.
[38] LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[C]//Proceedings of the IEEE. Piscataway:IEEE Press, 1998:2278-2324.
[39] NETZER Y, WANG T, COATES A, et al. Reading digits in natural images with unsupervised feature learning[C]//Proceeding of Neural Information Processing Systems Workshop, 2011.
[40] KRIZHEVSKY N, HINTON. The Cifar-10 Dataset[EB/OL].[2020-06-15]. http://www.cs.toronto.edu/kriz/cifar.html.
[41] DENG J, DONG W, SOCHER R, et al. Imagenet:A large-scale hierarchical image database[C]//Proceeding of IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2009:248-255.
[42] EVERINGHAM M, GOOL L V, WILLIAMS C K, et al. The pascal Visual Object Classes (VOC) challenge[J]. International Journal of Computer Vision, 2010, 88(2):303-338.
[43] LIN S Y, MAIRE M, BELONGIE S, et al. Microsoft COCO:Common objects in context[C]//Proceedings of the European Conference on Computer Vision, 2014:740-755.
[44] BULAT A, TZIMIROPOULOS G. XNOR-net++:Improved binary neural networks[C]//Proceeding of the British Machine Vision Conference, 2019.
[45] WANG Z W, LU J W, TAO C X, et al. Learning channel-wise interactions for binary convolutional neural networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2019:568-577.
[46] LIN X F, ZHAO C, PAN W. Towards accurate binary convolutional neural network[C]//Proceeding of the Conference of Advances in Neural Information Processing Systems, 2017.
[47] GEIGER A, LENZ P, STILLER C, et al. Vision meetsrobotics:The KITTI dataset[J]. The International Journal of Robotics Research, 2013, 32(11):1231-1237.
[48] 胡骏飞, 文志强, 谭海湖. 基于二值化卷积神经网络的手势分类方法研究[J]. 湖南工业大学学报, 2017, 31(1):75-80. HU J F, WEN Z Q, TAN H H. Research on gesture classification method based on binary convolutional neural network[J]. Journal of Hunan University of Technology, 2017, 31(1):75-80(in Chinese).
[49] FEIST T. Vivado design suite[EB/OL].[2020-06-15]. https://core.ac.uk/display/23362209.
[50] QIN H T, GONG R H, LIU X L, Bai X, et al. Binary neural networks:A survey[J]. Pattern Recognition, 2020, 105:107281.
[51] YANG H J, FRITZSCHE M, BARTZ C, et al. Bmxnet:An open-source binary neural network implementation based on MXNet[C]//Proceedings of the ACM Multimedia Conference, 2017.
[52] ZHAO T L, HE X Y, CHENG J, et al. Bitstream:Efficient computing architecture for real-time low-power inference of binary neural networks on CPUs[C]//Proceedings of the ACM Multimedia Conference, 2018:1545-1552.
[53] HU Y W, ZHAI J D, LI D H, et al. BitFlow:Exploiting vector parallelism for binary neural networks on CPU[C]//Proceeding of IEEE International Parallel and Distributed Processing Symposium. Piscataway:IEEE Press, 2018:244-253.
[54] ZHANG J H, PAN Y W, YAO T, et al. DABNN:A super fast inference framework for binary neural networks on ARM devices[C]//Proceeding of ACM Multimedia Conference. New York:ACM, 2019.
[55] GONG R H, LIU X L, JIANG S H, et al. Differentiable soft quantization:bridging full-precision and low-bit neural networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2019.

Options

文章导航

模态框（Modal）标题

摘要

本文引用格式

Abstract

参考文献