Deep convolutional neural network (CNN) is difficult to deploy to mobile and portable devices due to its large number of parameters and floating-point operations (FLOPs). To tackle this problem, we propose a novel channel pruning method. We use the modified squeeze-and-excitation blocks (MSEB) to measure the importance of the channels in the convolutional layers. The unimportant channels, including convolutional kernels related to them, are pruned directly, which greatly reduces the storage cost and the number of calculations. For ResNet with basic blocks, we propose an approach to consistently prune all residual blocks in the same stage to ensure that the compact network structure is dimensionally correct. After pruning we retrain the compact network from scratch to restore the accuracy. Finally, we verify our method on CIFAR-10, CIFAR-100 and ILSVRC-2012. The results indicate that the performance of the compact network is better than the original network when the pruning rate is small. Even when the pruning amplitude is large, the accuracy can also be maintained or decreased slightly. On the CIFAR-100, when reducing the parameters and FLOPs up to 82% and 62% respectively, the accuracy of VGG-19 even improve by 0.54% after retraining. The source code is available at https://github.com/JingfeiChang/UCP.
HintonG.E., OsinderoS. and TehY.-W., A fast learning algorithmfor deep belief nets, Neural Computation18(7) (2006), 1527–1554.
2.
HintonG.E. and SalakhutdinovR.R., Reducing the dimensionality ofdata with neural networks, Science313(5786) (2006), 504–507.
3.
LeCunY., BengioY. and HintonG., Deep learning, Nature521(7553) (2015), 436–444.
4.
LeCunY., KavukcuogluK. and FarabetC., “Convolutional networks and applications in vision,” in 2010 IEEE International Symposium on Circuits and Systems, 2010, pp. 253–256.
5.
LecunY., BottouL., BengioY. and HaffnerP., Gradientbasedlearning applied to document recognition, Proceedings of theIEEE86 (1998), 2278–324.
6.
KrizhevskyA., SutskeverI. and HintonG., Imagenet classificationwith deep convolutional neural networks, Communications of theACM60 (2017), 84–90.
7.
ZeilerM.D. and FergusR., “Visualizing and understandingconvolutional networks,” in, ECCV 20148689 (2014), 818–833.
8.
SimonyanK. and ZissermanA., “Very Deep Convolutional Networks for Large-Scale Image Recognition,” arXiv eprints, Sep 2014.
9.
SzegedyC., LiuW., JiaY., SermanetP., ReedS., AnguelovD., ErhanD., VanhouckeV. and RabinovichA., “Going deeper with convolutions,” in 2015 CVPR, 2015, pp. 1–9.
10.
HeK., ZhangX., RenS. and SunJ., “Deep residual learning for image recognition,” in CVPR, 2016, pp. 770–778.
11.
HuangGao, ZhuangLiu and WeinbergerK., “Densely connected convolutional networks,” arXiv, 24 Aug. 2016.
12.
HanS., PoolJ., TranJ. and DallyW.J., “Learning both weights and connections for efficient neural networks,” in NIPS 2015, 2015.
13.
LiH., KadavA., DurdanovicI., SametH. and GrafH.P., “Pruning Filters for Efficient ConvNets,” arXiv e-prints, Aug 2016.
14.
HuHengyuan, PengRui, TaiYu-Wing and TangChi-Keung, “Network Trimming: A Data-Driven Neuron Pruning Approach towards Efficient Deep Architectures [arXiv],” arXiv, p. 9 pp., 12 July 2016.
15.
LuoJ.-H. and WuJ., “An Entropy-based Pruning Method for CNN Compression,” arXiv e-prints, Jun 2017.
16.
YangT.-J., ChenY.-H. and SzeV., “Designing energy-efficient convolutional neural networks using energy-aware pruning,” in CVPR, 2017, pp. 6071–6079.
17.
HeY., KangG., DongX., FuY. and YangY., “Soft filter pruning for accelerating deep convolutional neural networks,” in IJCAI, 2018, pp. 2234–2240.
18.
HuangH., DengH., ShengY. and YeX., Accelerating convolutionalneural network-based malware traffic detection through ant-colonyclustering, Journal of Intelligent & Fuzzy Systems37(1) (2019), 409–423.
19.
LuoJ.-H., ZhangH., ZhouH.-Y., XieC.-W., WuJ. and LinW., “Thinet: Pruning cnn filters for a thinner net,”, IEEETranscations on Pattern Analysis and Machine Intelligence41(10) (2019), 2525–2538.
20.
AshouriA.H., AbdelrahmanT.S. and RemediosA.D., Retraining-freemethods for fast on-the-fly pruning of convolutional neuralnetworks, Neurocomputing370 (2019), 56–69.
21.
TooE.C., YujianL., KwaoP., NjukiS., MosomiM.E. and KibetJ., Deep pruned nets for efficient image-based plants diseaseclassification, Journal of Intelligent & Fuzzy Systems37(3) (2019), 4003–4019.
22.
LiuZ., LiJ., ShenZ., HuangG., YanS. and ZhangC., “Learning efficient convolutional networks through network slimming, in ICCV, 2017, pp. 2755–2763.
23.
HeY., ZhangX. and SunJ., “Channel pruning for accelerating very deep neural networks,” in ICCV, 2017, pp. 1398–1406.
24.
DingXiaohan, DingGuiguang, GuoYuchen and HanJungong, “Centripetal SGD for Pruning Very Deep Convolutional Networks With Complicated Structure,” in 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Proceedings, 2019 2019, Conference Paper, pp. 4938-48.
25.
TianQ., ArbelT. and ClarkJ.J., Structured deep fisher pruning forefficient facial trait classification, Image and VisionComputing77 (2018), 45–59.
26.
LinS., JiR., YanC., ZhangB., CaoL., YeQ., HuangF. and DoermannD.S., “Towards optimal structured cnn pruning via generative adversarial learning.” in CVPR, 2019, pp. 2790–2799.
27.
YouZ., YanK., YeJ., MaM. and WangP., “Gate Decorator: Global Filter Pruning Method for Accelerating Deep Convolutional Neural Networks,” arXiv e-prints, Sep 2019.
28.
SinghP., KadiV.S.R. and NamboodiriV.P., Falf convnets: Fatuousauxiliary loss based filter-pruning for efficient deep cnns, Imageand Vision Computing, 93 (2020), 103857.
29.
HewahiN.M., “Neural network pruning based on input importance,”, Journal of Intelligent & Fuzzy Systems37(2) (2019), 2243–2252.
30.
XieZ., ZhuL., ZhaoL., TaoB., LiuL. and TaoW., Localization-aware channel pruning for object detection, Neurocomputing, 2020.
31.
HuangY. and WangZ., Multi-granularity pruning for deep residualnetworks, Journal of Intelligent & Fuzzy Systems39(5) (2020), 7403–7410.
32.
LongX., ZengX., LiuY., XiaoH., ZhangM. and BenZ., A CNN channel pruning low-bit framework using weight quantization withsparse group lasso regularization, Journal of Intelligent &Fuzzy Systems39(1) (2020), 221–232.
33.
LuoJ.-H. and WuJ., Autopruner: An end-to-end trainable filterpruning method for efficient deep model inference, PatternRecognition107 (2020), 107461.
34.
HuJ., ShenL. and SunG., “Squeeze-and-excitation networks,” in CVPR, 2018, pp. 7132–7141.
35.
LiuZ., SunM., ZhouT., HuangG. and DarrellT., “Rethinking the value of network pruning,” in International Conference on Learning Representations, 2019.
36.
KammaK., IsodaY., InoueS. and WadaT., Neural behavior-basedapproach for neural network pruning, IEICE Transactions onInformation and SystemsE103D(5) (2020), 1135–1143.
37.
LongX., ZengX., ChenC., XiaoH. and ZhangM., Loss-driven channelpruning of convolutional neural networks, IEICE Transactions onInformation and SystemsE103D(5) (2020), 1190–1194.
38.
HeK., ZhangX., RenS. and SunJ., “Identity mappings in deep residual networks,” in ECCV (4) (2016), 630–645.
39.
LinM., JiR., ZhangY., ZhangB., WuY. and TianY., “Channel pruning via automatic structure search,” in Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI), 2020, pp. 673–679.
40.
YuR., LiA., ChenC., LaiJ., MorariuV.I., HanX., GaoM., LinandC. and DavisL.S., “Nisp: Pruning networks using neuron importance score propagation,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 9194–9203.
41.
MolchanovP., MallyaA., TyreeS., FrosioI. and KautzJ., “Importance estimation for neural network pruning,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019.
