Convolutional neural networks at constrained time cost
Open accessThe DOI matches its Crossref record; 26 of the 30 references it cites were found in a registry or an index.
Abstract
Though recent advanced convolutional neural networks (CNNs) have been improving the image recognition accuracy, the models are getting more complex and time-consuming. For real-world applications in industrial and commercial scenarios, engineers and developers are often faced with the requirement of constrained time budget. In this paper, we investigate the accuracy of CNNs under constrained time cost. Under this constraint, the designs of the network architectures should exhibit as trade-offs among the factors like depth, numbers of filters, filter sizes, etc. With a series of controlled comparisons, we progressively modify a baseline model while preserving its time complexity. This is also helpful for understanding the importance of the factors in network designs. We present an architecture that achieves very competitive accuracy in the ImageNet dataset (11.8% top-5 error, 10-view test), yet is 20% faster than “AlexNet” [14] (16.0% top-5 error, 10-view test).
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“The performance of neural network models is closely related to the setting of hyperparameters [42].”
From Prediction of Heat and Cold Loads of Factory Mushroom Houses Based on EWT Decomposition (2023) · cites this paper as [42]
Citations
Cites
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 9×
“No other data augmentation, such as scale jittering [10, 22, 9], is used.”
BackgroundMethodInfluentialImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 14×
“The actual running time of our model is 20% faster than AlexNet.”
BackgroundMethodResultInfluentialImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333 · cited 7×
“Through the investigations, we obtain a model that achieves 11.8% top-5 error (10-view test) on ImageNet [3] and only takes 3 to 4 days training on a single GPU.”
BackgroundMethodInfluentialGoing deeper with convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., & Rabinovich, A. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298594 · Cited by 48,562
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380
Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation
Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014) · 2014 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2014.81 · Cited by 32,684
(2010) · Cited by 14,134
(2010) · International Conference on Machine Learning · Cited by 13,995
Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2015) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2015.2389824 · Cited by 11,971
Self-citationBackpropagation Applied to Handwritten Zip Code Recognition
LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., & Jackel, L. D. (1989) · Neural Computation · doi:10.1162/neco.1989.1.4.541 · Cited by 11,891 · cited 1×
“Convolutional neural networks (CNNs) [15, 14] have recently brought in revolutions to the computer vision area.”
Background
Research works printed but not linked: 1
- lin, 2013, Network in network, arXiv 1312 4400
Cited by
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474
“Unexpectedly, such degradation is not caused by overfitting, and adding more layers to a suitably deep model leads to higher training error, as reported in [11, 42] and thoroughly verified by our experiments.”
ResultSelf-citationDelving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification
He, K., Zhang, X., Ren, S., & Sun, J. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.123 · Cited by 19,741
“In [11], aggressively increasing the depth leads to saturated or degraded accuracy.”
BackgroundMethodInfluentialSelf-citationImage Super-Resolution Using Deep Convolutional Networks
Dong, C., Loy, C. C., He, K., & Tang, X. (2016) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2015.2439281 · Cited by 10,105
Self-citationShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices
Zhang, X., Zhou, X., Lin, M., & Sun, J. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00716 · Cited by 9,982
Self-citationShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices
Zhang, X., Zhou, X., Lin, M., & Sun, J. (2017) · arXiv · doi:10.48550/arxiv.1707.01083 · Cited by 9,982
“The increasing needs of running high quality deep neural networks on embedded devices encourage the study on efficient model designs [8].”
BackgroundSelf-citationSqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size
Iandola, F. N., Han, S., Moskewicz, M. W., Ashraf, K., Dally, W. J., & Keutzer, K. (2016) · arXiv · doi:10.48550/arxiv.1602.07360 · Cited by 6,201
“Sun applied delayed downsampling to four different DNN architectures, and in each case delayed downsampling led to higher classification accuracy [10].”
MethodDeep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2015) · arXiv · doi:10.48550/arxiv.1512.03385 · Cited by 4,196
“Unexpectedly, such degradation is not caused by overfitting, and adding more layers to a suitably deep model leads to higher training error, as reported in [11, 42] and thoroughly verified by our experiments.”
ResultSelf-citationObject Detection in 20 Years: A Survey
Zou, Z., Chen, K., Shi, Z., Guo, Y., & Ye, J. (2023) · Proceedings of the IEEE · doi:10.1109/jproc.2023.3238524 · Cited by 3,689
Deep Convolutional Neural Networks for Image Classification: A Comprehensive Review
Rawat, W., & Wang, Z. (2017) · Neural Computation · doi:10.1162/neco_a_00990 · Cited by 3,503
“Notwithstanding degradation (He et al., 2015b; He & Sun, 2015; Srivastava et al., 2015a, 2015b), deeper models are generally more accurate and thus produce better empirical results; however, as depth increases, so do computational costs.”
BackgroundMethodInfluentialPruning Filters for Efficient ConvNets
Li, H., Kadav, A., Durdanovic, I., Samet, H., & Graf, H. P. (2016) · arXiv · doi:10.48550/arxiv.1608.08710 · Cited by 2,928
“Additionally, quantization (Han et al. (2016b)) and binarization (Rastegari et al. (2016); Courbariaux & Bengio (2016)) can be used to reduce the model size and lower the computation overheads.”
BackgroundMethodDeep learning for visual understanding: A review
Guo, Y., Liu, Y., Oerlemans, A., Lao, S., Wu, S., & Lew, M. S. (2016) · Neurocomputing · doi:10.1016/j.neucom.2015.09.116 · Cited by 2,504
“Nevertheless, there are approaches [18,27,146] that deliver better performance by training on other models, e.”
BackgroundA State-of-the-Art Survey on Deep Learning Theory and Architectures
Alom, M. Z., Taha, T. M., Yakopcic, C., Westberg, S., Sidike, P., Nasrin, M. S., Hasan, M., Van Essen, B. C., Awwal, A. A. S., & Asari, V. K. (2019) · Electronics · doi:10.3390/electronics8030292 · Cited by 1,620
Predicting residential energy consumption using CNN-LSTM neural networks
Kim, T.-Y., & Cho, S.-B. (2019) · Energy · doi:10.1016/j.energy.2019.05.230 · Cited by 1,469
“Adjusting these parameters can affect learning speed and performance depending on the characteristics of the learning data [44].”
BackgroundDynamic Edge-Conditioned Filters in Convolutional Neural Networks on Graphs
Simonovsky, M., & Komodakis, N. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.11 · Cited by 1,266
“Therefore, the restriction of ECC to 1-hop neighborhoods N(i) is not a constraint, akin to using small 3×3 filters in normal CNNs in exchange for deeper networks, which is known to be beneficial [17].”
Accelerating Very Deep Convolutional Networks for Classification and Detection
Zhang, X., Zou, J., He, K., & Sun, J. (2016) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2015.2502579 · Cited by 832
“It may not be beneficial to uniformly accelerate all layers.”
BackgroundSelf-citationDeep Learning in Medical Ultrasound Analysis: A Review
Liu, S., Wang, Y., Yang, X., Lei, B., Liu, L., Li, S. X., Ni, D., & Wang, T. (2019) · Engineering · doi:10.1016/j.eng.2018.11.020 · Cited by 798
“In practice, smaller strides always work better because small strides in the early layers of the network (i.e., those layers that are closer to the input data) can generate a large activation map, which can lead to better performance [56].”
BackgroundPredicting Human Eye Fixations via an LSTM-Based Saliency Attentive Model
Cornia, M., Baraldi, L., Serra, G., & Cucchiara, R. (2018) · IEEE Transactions on Image Processing · doi:10.1109/tip.2018.2851672 · Cited by 474
Attention Residual Learning for Skin Lesion Classification
Zhang, J., Xie, Y., Xia, Y., & Shen, C. (2019) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2019.2893944 · Cited by 568
“However, due to degradation problems [41], [42], it is still”
BackgroundA Survey on Machine Learning Techniques for Cyber Security in the Last Decade
Shaukat, K., Luo, S., Varadharajan, V., Hameed, I. A., & Xu, M. (2020) · IEEE Access · doi:10.1109/access.2020.3041951 · Cited by 530
Learning-Based Computation Offloading for IoT Devices With Energy Harvesting
Min, M., Xiao, L., Chen, Y., Cheng, P., Wu, D., & Zhuang, W. (2019) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2018.2890685 · Cited by 560
Yang, X., Sun, H., Fu, K., Yang, J., Sun, X., Yan, M., & Guo, Z. (2018) · Remote Sensing · doi:10.3390/rs10010132 · Cited by 546
“With the application of deep convolutional networks [9-15] in object detection, more and more efficient detection algorithms are proposed, such as region proposals with CNNs (RCNN) [16], Spatial Pyramid Pooling Network (SSP-Net) [17] and Fast-RCNN [18].”
MethodPractical Block-Wise Neural Network Architecture Generation
Zhong, Z., Yan, J., Wu, W., Shao, J., & Liu, C.-L. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00257 · Cited by 522
“where FLOPs [8] refer to an estimation of computational complexity of the block, and Density is the edge number divided by the dot number in DAG of the block.”
BackgroundInfluentialYou, C., Cong, W., Vannier, M. W., Saha, P. K., Hoffman, E. A., Wang, G., Li, G., Zhang, Y., Zhang, X., Shan, H., et al. (2020) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2019.2922960 · Cited by 459
“2) Number of Layers: Recent studies reported in [72], [73] suggest that training a network could benefit from increasing the network depth moderately.”
BackgroundINet: Convolutional Networks for Biomedical Image Segmentation
Weng, W., & Zhu, X. (2021) · IEEE Access · doi:10.1109/access.2021.3053408 · Cited by 509
“The computational complexity of a single Conv-layer [55] can be measured by: where M is the side length of each convolution kernel’s output feature map.”
BackgroundCompressed residual-VGG16 CNN model for big data places image recognition
Qassim, H., Verma, A., & Feinzimer, D. (2018) · 2018 IEEE 8th Annual Computing and Communication Workshop and Conference (CCWC) · doi:10.1109/ccwc.2018.8301729 · Cited by 486
Nonparametric regression using deep neural networks with ReLU activation function
Schmidt-Hieber, J. (2020) · The Annals of Statistics · doi:10.1214/19-aos1875 · Cited by 292
Liu, R., Wang, F., Yang, B., & Qin, S. J. (2020) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2019.2941868 · Cited by 454
Deep CNN-Based Channel Estimation for mmWave Massive MIMO Systems
Dong, P., Zhang, H., Li, G. Y., Gaspar, I. S., & NaderiAlizadeh, N. (2019) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2019.2925975 · Cited by 437
“According to [25], the required FLOPs of SF-CNN processing is CSF-CNN ∼ O(∑Lc l=1 M1,lM2,lF 2 l Nl−1Nl), where Lc is number of convolutional layers, M1,l and M2,l denote the number of rows and columns of each feature map output by the lth layer, Fl is the side length of the filters used by the lth layer, Nl−1 and Nl d…”
MethodScene Classification via a Gradient Boosting Random Convolutional Network Framework
Zhang, F., Du, B., & Zhang, L. (2016) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2015.2488681 · Cited by 424
“[31] also suggests that CNet can extract a more efficient high-level feature from the deeper network than the shallow network.”
BackgroundApplications of Deep Learning to Ocean Data Inference and Subgrid Parameterization
Bolton, T., & Zanna, L. (2019) · Journal of Advances in Modeling Earth Systems · doi:10.1029/2018ms001472 · Cited by 399
Residual Networks of Residual Networks: Multilevel Residual Networks
Zhang, K., Sun, M., Han, T. X., Yuan, X., Guo, L., & Liu, T. (2018) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2017.2654543 · Cited by 401
Fast ConvNets Using Group-Wise Brain Damage
Lebedev, V., & Lempitsky, V. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.280 · Cited by 373
“It is interesting to see that the optimization process decided to shrink the receptive fields towards the center compared to the full version (which is consistent with the findings in [43, 19]).”
BackgroundResultA review of convolutional neural network architectures and their optimizations
Cong, S., & Zhou, Y. (2023) · Artificial Intelligence Review · doi:10.1007/s10462-022-10213-5 · Cited by 381
“The error rate arising from such problem is attributed merely to the increased number of network layers, rather than overfitting (He and Sun 2015).”
BackgroundSabokrou, M., Fayyaz, M., Fathy, M., & Klette, R. (2017) · IEEE Transactions on Image Processing · doi:10.1109/tip.2017.2670780 · Cited by 386
Deep Residual Learning in Spiking Neural Networks
Fang, W., Yu, Z., Chen, Y., Huang, T., Masquelier, T., & Tian, Y. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.04159 · Cited by 209
“Nevertheless, recent evidence [13, 53, 14] reveals that with the network depth increasing, the accuracy gets saturated and then degrades rapidly.”
BackgroundA Deep Learning Framework for Optimization of MISO Downlink Beamforming
Xia, W., Zheng, G., Zhu, Y., Zhang, J., Wang, J., & Petropulu, A. P. (2020) · IEEE Transactions on Communications · doi:10.1109/tcomm.2019.2960361 · Cited by 345
“Thus, the total time complexity of all convolutional layers measured by the number of multiplications is l c l − 1 c l [52].”
BackgroundWeb traffic anomaly detection using C-LSTM neural networks
Kim, T.-Y., & Cho, S.-B. (2018) · Expert Systems with Applications · doi:10.1016/j.eswa.2018.04.004 · Cited by 340
Liu, C., Cao, Y., Luo, Y., Chen, G., Vokkarane, V., Yunsheng, M., Chen, S., & Hou, P. (2018) · IEEE Transactions on Services Computing · doi:10.1109/tsc.2017.2662008 · Cited by 333
“layers, we resort to the lessons learned in recent paper [50],”
MethodIntelligent Mechanical Fault Diagnosis Using Multisensor Fusion and Convolution Neural Network
Xie, T., Huang, X., & Choi, S.-K. (2022) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2021.3102017 · Cited by 306
A Densely Connected End-to-End Neural Network for Multiscale and Multiscene SAR Ship Detection
Jiao, J., Zhang, Y., Sun, H., Yang, X., Gao, X., Hong, W., Fu, K., & Sun, X. (2018) · IEEE Access · doi:10.1109/access.2018.2825376 · Cited by 299
“With the development of computer hardware and deep learning, convolutional neural network (CNN) has become the dominate approach for object detection, classification, and segmentation [14]–[17].”
BackgroundDeepFood: Deep Learning-Based Food Image Recognition for Computer-Aided Dietary Assessment
Liu, C., Cao, Y., Luo, Y., Chen, G., Vokkarane, V., & Ma, Y. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-39601-9_4 · Cited by 289
“In [38], Kaiming et al. gives the general guidance for modifying models considering the influence of different depth, number of filters and filter size.”
BackgroundMachine Learning Empowered Trajectory and Passive Beamforming Design in UAV-RIS Wireless Networks
Liu, X., Liu, Y., & Chen, Y. (2021) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2020.3041401 · Cited by 298
“where i denotes the number of Conv layer [39].”
Background(2021) · doi:10.1109/twc.2021.3100148 · Cited by 290
“As for the proposed method, its computational cost is dominated by the computations of the CDRN [54] and the”
MethodDomain Adversarial Graph Convolutional Network for Fault Diagnosis Under Variable Working Conditions
Li, T., Zhao, Z., Sun, C., Yan, R., & Chen, X. (2021) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2021.3075016 · Cited by 280
“D convolutional (Conv) layer is given by [44].”
BackgroundA Framework for Designing the Architectures of Deep Convolutional Neural Networks
Albelwi, S., & Mahmood, A. (2017) · Entropy · doi:10.3390/e19060242 · Cited by 278
“[15,16] showed that increasing the depth harmed the performance, as further proven by the experiments in [17].”
ResultDeep CM-CNN for Spectrum Sensing in Cognitive Radio
Liu, C., Wang, J., Liu, X., & Liang, Y.-C. (2019) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2019.2933892 · Cited by 280
Learning-Based Predictive Beamforming for Integrated Sensing and Communication in Vehicular Networks
Liu, C., Yuan, W., Li, S., Liu, X., Li, H., Ng, D. W. K., & Li, Y. (2022) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2022.3180803 · Cited by 264
“According to [44], the computation of a CNN module at each time step is Here, O ( · ) represents the order of the computational complexity, N L represents the number of convolutional layers, and n 0 is the input dimension.”
MethodFully Convolutional Architectures for Multiclass Segmentation in Chest Radiographs
Novikov, A. A., Lenis, D., Major, D., Hladuvka, J., Wimmer, M., & Buhler, K. (2018) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2018.2806086 · Cited by 252
MFQE 2.0: A New Approach for Multi-Frame Quality Enhancement on Compressed Video
Guan, Z., Xing, Q., Xu, M., Yang, R., Liu, T., & Wang, Z. (2021) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2019.2944806 · Cited by 259
“apply the dense strategy and then decrease the number of output feature maps. This is because when the time cost is limited, the layer width is of less importance than the network depth, as proven in [54]. This way, the effectiveness of our QE-subnet can be greatly enhanced, while its complexity remains at a low level…”
MethodA hybrid deep learning CNN–ELM for age and gender classification
Duan, M., Li, K., Yang, C., & Li, K. (2018) · Neurocomputing · doi:10.1016/j.neucom.2017.08.062 · Cited by 258
PolyNet: A Pursuit of Structural Diversity in Very Deep Networks
Zhang, X., Li, Z., Loy, C. C., & Lin, D. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.415 · Cited by 249
“In addition, extensive studies [6, 7, 26, 17] have been done to find good design choices, e.”
BackgroundDeepOPF: A Deep Neural Network Approach for Security-Constrained DC Optimal Power Flow
Pan, X., Zhao, T., Chen, M., & Zhang, S. (2021) · IEEE Transactions on Power Systems · doi:10.1109/tpwrs.2020.3026379 · Cited by 252
“ly generate infeasible solutions, as seen in Sec. VI. Thus, the computational complexity of DeepOPF is dominated by the calculation with respect to the DNN model. It can be evaluated by the method in [41]. Specifically, recall that the number of bus and the number of contingencies are Nbus and n c, respectively. The in…”
MethodTsironi, E., Barros, P., Weber, C., & Wermter, S. (2017) · Neurocomputing · doi:10.1016/j.neucom.2016.12.088 · Cited by 248
“The complexity of all convolutional layers [21] is estimated as O(Σl=1nl−1 · s(2)l · nl ·m(2)l ), where d is the number of convolutional layers, nl is the number of filters in the l-th layer, nl−1 is the number of input channels of the l layer, sl is the spatial size of the filter, and ml is 165”
BackgroundA Survey on Methods and Theories of Quantized Neural Networks
Guo, Y. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1808.04752 · Cited by 188
Haase, D., & Amthor, M. (2020) · 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr42600.2020.01461 · Cited by 220
SAR Image segmentation based on convolutional-wavelet neural network and markov random field
Duan, Y., Liu, F., Jiao, L., Zhao, P., & Zhang, L. (2017) · Pattern Recognition · doi:10.1016/j.patcog.2016.11.015 · Cited by 244
Yi, Y., Zhang, Z., Zhang, W., Zhang, C., Li, W., & Zhao, T. (2019) · Remote Sensing · doi:10.3390/rs11151774 · Cited by 225
“However, for experimental research and practical application, the trade-offs among the layer depth, number of channels, kernel sizes, and other attributes of the network must still be considered when designing the architectures of networks [61], given the concern of cost-effectiveness in training time and commercial c…”
BackgroundDeep-Learning-Based Open Set Fault Diagnosis by Extreme Value Theory
Yu, X., Zhao, Z., Zhang, X., Zhang, Q., Liu, Y., Sun, C., & Chen, X. (2022) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2021.3070324 · Cited by 222
What Your Images Reveal: Exploiting Visual Contents for Point-of-Interest Recommendation
Wang, S., Wang, Y., Tang, J., Shu, K., Ranganath, S., & Liu, H. (2017) · Proceedings of the 26th International Conference on World Wide Web · doi:10.1145/3038912.3052638 · Cited by 229
“Then updating filter weights of the lth convolutional layer for one input costs O(fls(2)l nlm(2)l ) [12].”
MethodUllah, F. U. M., Ullah, A., Haq, I. U., Rho, S., & Baik, S. W. (2020) · IEEE Access · doi:10.1109/access.2019.2963045 · Cited by 222
“This setting affects the rate of learning and the network performance, which depend on the nature of the learning data [52].”
BackgroundStrawberry Yield Prediction Based on a Deep Neural Network Using High-Resolution Aerial Orthoimages
Chen, Y., Lee, W. S., Gan, H., Peres, N., Fraisse, C., Zhang, Y., & He, Y. (2019) · Remote Sensing · doi:10.3390/rs11131584 · Cited by 224
“As shown in Figure 5, instead of stacking layers directly to fit a desired underlying mapping, like VGG nets [51], ResNet-50 introduces a deep residual learning framework to fit a residual mapping which helps to address the degradation problem [52].”
BackgroundAbrol, A., Fu, Z., Salman, M., Silva, R., Du, Y., Plis, S., & Calhoun, V. (2021) · Nature Communications · doi:10.1038/s41467-020-20655-6 · Cited by 221
D3: Deep Dual-Domain Based Fast Restoration of JPEG-Compressed Images
Wang, Z., Liu, D., Chang, S., Ling, Q., Yang, Y., & Huang, T. S. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.302 · Cited by 215
Deep Learning-Based Spectrum Sensing in Cognitive Radio: A CNN-LSTM Approach
Xie, J., Fang, J., Liu, C., & Li, X. (2020) · IEEE Communications Letters · doi:10.1109/lcomm.2020.3002073 · Cited by 211
“In the network training process, for CNN, accroding to [17], its computational complexity for one data sample is O ( (cid:5) , where, I is the number of convolutional layers, n c,i is the number of convolutional kernels in the i th layer and n c,i − 1 can also be known as the number of input…”
BackgroundResidual Attention U-Net for Automated Multi-Class Segmentation of COVID-19 Chest CT Images
Chen, X., Yao, L., & Zhang, Y. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.05645 · Cited by 177
“This problem is be defined as degradation problem [25], [26].”
BackgroundIdentifying critical nodes in complex networks via graph convolutional networks
Yu, E.-Y., Wang, Y.-P., Fu, Y., Chen, D.-B., & Xie, M. (2020) · Knowledge-Based Systems · doi:10.1016/j.knosys.2020.105893 · Cited by 211
MODE: automated neural network model debugging via state differential analysis and input selection
Ma, S., Liu, Y., Lee, W.-C., Zhang, X., & Grama, A. (2018) · Proceedings of the 2018 26th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering · doi:10.1145/3236024.3236082 · Cited by 211
Clockwork Convnets for Video Semantic Segmentation
Shelhamer, E., Rakelly, K., Hoffman, J., & Darrell, T. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-49409-8_69 · Cited by 197
“Careful but time-consuming model search can improve networks within a fixed computational budget [20].”
BackgroundLiu, X., Liu, Y., Chen, Y., & Poor, H. V. (2021) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2020.3018823 · Cited by 208
“As demonstrated in [51], the CNN’s computational complexity is given by O(Ξ) =”
BackgroundHuang, Y.-J., Liao, A.-H., Hu, D.-Y., Shi, W., & Zheng, S.-B. (2022) · Measurement · doi:10.1016/j.measurement.2022.111935 · Cited by 197
Attention Mask R-CNN for ship detection and segmentation from remote sensing images
Nie, X., Duan, M., Ding, H., Hu, B., & Wong, E. K. (2020) · IEEE Access · doi:10.1109/access.2020.2964540 · Cited by 200
“In recent years, CNN (convolutional neural network) has become the dominant method for image classification, target detection and segmentation [5]–[7].”
MethodKim, J., Kim, J., Jang, G.-J., & Lee, M. (2017) · Neural Networks · doi:10.1016/j.neunet.2016.12.002 · Cited by 199
RFAL: Adversarial Learning for RF Transmitter Identification and Classification
Roy, D., Mukherjee, T., Chatterjee, M., Blasch, E., & Pasiliao, E. (2020) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2019.2948919 · Cited by 192
“In [56], the mentioned that the time complexity for training all the convolutional layers is: (cid”
BackgroundMethodA bearing fault diagnosis model based on CNN with wide convolution kernels
Song, X., Cong, Y., Song, Y., Chen, Y., & Liang, P. (2022) · Journal of Ambient Intelligence and Humanized Computing · doi:10.1007/s12652-021-03177-x · Cited by 183
“These layers usually take 5-10% of the calculation time (He and Sun 2015).”
BackgroundDistance metric optimization driven convolutional neural network for age invariant face recognition
Li, Y., Wang, G., Nie, L., Wang, Q., & Tan, W. (2018) · Pattern Recognition · doi:10.1016/j.patcog.2017.10.015 · Cited by 192
Differential Deep Convolutional Neural Network Model for Brain Tumor Classification
Abd El Kader, I., Xu, G., Shuai, Z., Saminu, S., Javaid, I., & Salim Ahmad, I. (2021) · Brain Sciences · doi:10.3390/brainsci11030352 · Cited by 179
“CNN deep networks have achieved great success in recent years in analyzing medical images, as a relatively low-cost means of detecting and classifying accuracy in contrast to its expensive computational costs that restrict the application of CNN in clinical practice [50].”
BackgroundYu, B., Yang, L., & Chen, F. (2018) · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · doi:10.1109/jstars.2018.2860989 · Cited by 182
“Adding more layers to a network model may lead to higher training error [47].”
BackgroundAn adaptive spatiotemporal feature learning approach for fault diagnosis in complex systems
Han, T., Liu, C., Wu, L., Sarkar, S., & Jiang, D. (2019) · Mechanical Systems and Signal Processing · doi:10.1016/j.ymssp.2018.07.048 · Cited by 181
Trends in Heart-Rate Variability Signal Analysis
Ishaque, S., Khan, N., & Krishnan, S. (2021) · Frontiers in Digital Health · doi:10.3389/fdgth.2021.639444 · Cited by 161
“…P , where l represents the index of the convolution layer, d represents the depth, n l represents the number of filters in the l -th layer, n l − 1 describes the number of input channels, s l indicates the spatial size of the filter and m l represents the spatial size of the output feature map (91).”
BackgroundRAMP-CNN: A Novel Neural Network for Enhanced Automotive Radar Object Recognition
Gao, X., Xing, G., Roy, S., & Liu, H. (2021) · IEEE Sensors Journal · doi:10.1109/jsen.2020.3036047 · Cited by 171
Lu, W., Tong, Y., Yu, Y., Xing, Y., Chen, C., & Shen, Y. (2018) · Translational Vision Science & Technology · doi:10.1167/tvst.7.6.41 · Cited by 171
“However, with the network depth increasing, accuracy gets saturated and then degrades rapidly.(20) The ResNet framework can tackle this problem.”
BackgroundSpecific Emitter Identification Based on Deep Residual Networks
Pan, Y., Yang, S., Peng, H., Li, T., & Wang, W. (2019) · IEEE Access · doi:10.1109/access.2019.2913759 · Cited by 167
“One obstacle is the notorious problem of vanishing/exploding gradients [26], which hampers convergence from the beginning.”
BackgroundKensert, A., Harrison, P. J., & Spjuth, O. (2019) · SLAS Discovery · doi:10.1177/2472555218818756 · Cited by 165
Space-Time Correspondence as a Contrastive Random Walk
Jabri, A., Owens, A., & Efros, A. A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.14613 · Cited by 109
“We compare to a variety of methods, all of which use ResNet-18 [32] as the network backbone.”
ResultYang, X., Sun, H., Sun, X., Yan, M., Guo, Z., & Fu, K. (2018) · IEEE Access · doi:10.1109/access.2018.2869884 · Cited by 164
“In recent years, deep learning has achieved great success in computer vision [19]–[24].”
BackgroundLi, Y., Wu, J., & Wu, Q. (2019) · IEEE Access · doi:10.1109/access.2019.2898044 · Cited by 159
“The deep residual learning framework (ResNet) is proposed by He and Sun [31] to address the degradation of deep networks.”
BackgroundInfluentialReinforcement Learning-Based Mobile Offloading for Edge Computing Against Jamming and Interference
Xiao, L., Lu, X., Xu, T., Wan, X., Ji, W., & Zhang, Y. (2020) · IEEE Transactions on Communications · doi:10.1109/tcomm.2020.3007742 · Cited by 160
Moving Convolutional Neural Networks to Embedded Systems: The AlexNet and VGG-16 Case
Alippi, C., Disabato, S., & Roveri, M. (2018) · 2018 17th ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN) · doi:10.1109/ipsn.2018.00049 · Cited by 161
“As suggested in [24], the computational load ofΦ can be approximated as the sum of the computational load of the convolutional filterC conv and the one of the fully connected C f c layers, i.”
BackgroundA novel end-to-end 1D-ResCNN model to remove artifact from EEG signals
Sun, W., Su, Y., Wu, X., & Wu, X. (2020) · Neurocomputing · doi:10.1016/j.neucom.2020.04.029 · Cited by 170
Xie, D., Xiong, J., & Pu, S. (2017) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2017.539 · Cited by 206
“Moreover, we try to show that the degradation problem of training a plain network reported in [11, 10, 35] can be partially solved by our methods.”
MethodA Graph-Based Semisupervised Deep Learning Model for PolSAR Image Classification
Bi, H., Sun, J. Z., & Xu, Z. (2018) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2018.2871504 · Cited by 157
Self-citationDA-DCGAN: An Effective Methodology for DC Series Arc Fault Diagnosis in Photovoltaic Systems
Lu, S., Sirojan, T., Phung, B. T., Zhang, D., & Eliathamby Ambikairajah, A. (2019) · IEEE Access · doi:10.1109/access.2019.2909267 · Cited by 171
Sentiment analysis through recurrent variants latterly on convolutional neural network of Twitter
Abid, F., Alam, M., Yasir, M., & Li, C. (2019) · Future Generation Computer Systems · doi:10.1016/j.future.2018.12.018 · Cited by 151
“However, computational cost with time complexity in the aforementioned expression instead of the actual running time relies upon the presented architecture, due to its affectability which is depended on the elements of cost of implementations and particularly with the hardware by [61].”
BackgroundHyperspectral image classification using CNN with spectral and spatial features integration
Vaddi, R., & Manoharan, P. (2020) · Infrared Physics & Technology · doi:10.1016/j.infrared.2020.103296 · Cited by 157
Chu, H.-H., Wang, W., & Deng, L. (2022) · Computer-Aided Civil and Infrastructure Engineering · doi:10.1111/mice.12881 · Cited by 161
“Stacking multiple shallow subnetworks can improve the feature extraction ability of the network to some extent, while simply increasing the complexity or depth of the CNN architecture usually leads to degradation in accuracy (He & Sun, 2015; Srivastava et al., 2015).”
BackgroundKundu, N. K., & McKay, M. R. (2021) · IEEE Open Journal of the Communications Society · doi:10.1109/ojcoms.2021.3063171 · Cited by 175
“For a general CNN architecture, the computational complexity during online implementation is given by & [44], where L is the total number of convolutional layers, h l is the height and w l is width of the output feature map of the l − th layer, F l is the filter size in the l − the layer, and N l −…”
BackgroundPerformance Evaluation of Convolutional Neural Network for Hand Gesture Recognition Using EMG
Asif, A. R., Waris, A., Gilani, S. O., Jamil, M., Ashraf, H., Shafique, M., & Niazi, I. K. (2020) · Sensors · doi:10.3390/s20061642 · Cited by 150
“testing time per image due to one forward and two backward propagations [30].”
BackgroundA Survey on Large-Scale Machine Learning
Wang, M., Fu, W., He, X., Hao, S., & Wu, X. (2020) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2020.3015777 · Cited by 215
“However, to maintain the effectiveness with filter decomposition, it is essential to analyze the complexity of each operation carefully and modify a model progressively [49], [92].”
BackgroundHoang, L., Lee, S., Lee, E.-J., & Kwon, K. (2022) · Applied Sciences · doi:10.3390/app12052677 · Cited by 167
“The growing demand for streaming high-quality DL architectures on embedded systems boosts the research of effective model designs [62].”
BackgroundUAV-Aided Cellular Communications with Deep Reinforcement Learning Against Jamming
Lu, X., Xiao, L., Dai, C., & Dai, H. (2020) · IEEE Wireless Communications · doi:10.1109/mwc.001.1900207 · Cited by 145
“mming strategy chooses the maximum jamming power to attack the UAV-aided cellular system. The complexity of the deep RL based relay scheme as shown in Fig. 2 depends on the CNN complexity. Similar to [15], the CNN complexity depends on the number of input channels for the CNN, the numberof the two Conv layer filters f1…”
Background
Metadata
- Authors
- He, K., & Sun, J.
- Affiliations
- Microsoft (United States) (US)
- ORCID
- Venue
- 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
- Publisher
- IEEE
- Published
- 2015-06
- Pages
- 5353-5360
- ISSN
- 1063-6919 (ISSN-L)
- OpenAlex
- W1932847118
- Cited by
- 1,558
- References
- 30
- Citations by year
- 2026: 32 · 2025: 91 · 2024: 143 · 2023: 174 · 2022: 216 · 2021: 252 · 2020: 227 · 2019: 209 · 2018: 110 · 2017: 61 · 2016: 32 · 2015: 11
- FWCI
- 29.25
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Image and Video Retrieval Techniques
- Keywords
- convolutional neural networks, time complexity, inference speed, network architecture design, speed-accuracy trade-off, ImageNet classification, AlexNet, model compression
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.