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
Items 101–200 of 1,001
GAN augmentation to deal with imbalance in imaging-based intrusion detection
Andresini, G., Appice, A., Rose, L., & Malerba, D. (2021) · Future Generation Computer Systems · doi:10.1016/j.future.2021.04.017 · Cited by 162
Energy efficient ECG classification with spiking neural network
Yan, Z., Zhou, J. R., & Wong, W. (2020) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2020.102170 · Cited by 152
BlockQNN: Efficient Block-Wise Neural Network Architecture Generation
Zhong, Z., Yang, Z., Deng, B., Yan, J., Wu, W., Shao, J., & Liu, C. (2020) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2020.2969193 · Cited by 163
“Meanwhile, we notice that the FLOPs and density of the corresponding blocks have a negative correlation with the final accuracy.”
BackgroundInfluentialPerformance Analysis of GPU-Based Convolutional Neural Networks
Li, X., Zhang, G., Huang, H. H., Wang, Z., & Zheng, W. (2016) · 2016 45th International Conference on Parallel Processing (ICPP) · doi:10.1109/icpp.2016.15 · Cited by 136
“Especially for modern advanced CNN models, the computing cost of convolutional layers is getting much higher due to the increasingly more filters and layers, smaller strides and their combinations [17].”
BackgroundChen, S., Qin, J., Ji, X., Lei, B., Wang, T., Ni, D., & Cheng, J.-Z. (2017) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2016.2629462 · Cited by 142
Hybrid Quantum-Classical Convolutional Neural Network Model for Image Classification
Fan, F., Shi, Y., Guggemos, T., & Zhu, X. X. (2023) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2023.3312170 · Cited by 190
“As studied in [69], a classical convolution layer with the same settings as our QC-CNN model requires O ( 2 k + 2 n ) operations to process a gray-scale image of size 2 n × 2 n .”
Energy-Efficient Distributed Mobile Crowd Sensing: A Deep Learning Approach
Liu, C. H., Chen, Z., & Zhan, Y. (2019) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2019.2904353 · Cited by 147
“According to [35], the time complexity of CNN is O( ∑d l=1 s 2 l · m(2)l · nl−1 · nl), where l is the index of a CNN layer and d is the number of CNN layers, and ml and sl are the side lengths of the layer l’s output feature map and kernel, while nl is the number of CNN channels at CNN layer l.”
BackgroundTwo-Dimensional Antijamming Mobile Communication Based on Reinforcement Learning
Xiao, L., Jiang, D., Xu, D., Zhu, H., Zhang, Y., & Poor, H. V. (2018) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2018.2856854 · Cited by 139
Detection and classification of chilli leaf disease using a squeeze-and-excitation-based CNN model
Naik, B. N., Malmathanraj, R., & Palanisamy, P. (2022) · Ecological Informatics · doi:10.1016/j.ecoinf.2022.101663 · Cited by 139
Advanced agricultural disease image recognition technologies: A review
Yuan, Y., Chen, L., Wu, H., & Li, L. (2021) · Information Processing in Agriculture · doi:10.1016/j.inpa.2021.01.003 · Cited by 196
Xiang, Z., & Demir, İ. (2020) · Environmental Modelling & Software · doi:10.1016/j.envsoft.2020.104761 · Cited by 155
Niu, G., Liu, E., Wang, X., Ziehl, P. H., & Zhang, B. (2022) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2022.3179011 · Cited by 142
“saturates and degrades when the depth of the network exceeds a certain value [15], [24], [25].”
BackgroundLiu, C. H., Dai, Z., Zhao, Y., Crowcroft, J., Wu, D. O., & Leung, K. K. (2019) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2019.2938509 · Cited by 142
“According to [38], the time complexity of a CNN is O ð where l is the layer index, and L is the number of layers, and k l ; a l ; n l (cid:4) 1 ; n l are kernel size, output feature map size, input channel and output channel, respectively.”
BackgroundSeismic trace interpolation for irregularly spatial sampled data using convolutional autoencoder
Wang, Y., Wang, B., Tu, N., & Geng, J. (2019) · Geophysics · doi:10.1190/geo2018-0699.1 · Cited by 141
Recent Development on Detection Methods for the Diagnosis of Diabetic Retinopathy
Qureshi, I., Ma, J., & Abbas, Q. (2019) · Symmetry · doi:10.3390/sym11060749 · Cited by 135
“[157] presented a real-time applicable DL architecture by reducing the filter size and running time.”
MethodReal-Time Speech Emotion and Sentiment Recognition for Interactive Dialogue Systems
Bertero, D., Siddique, F. B., Wu, C.-S., Wan, Y., Chan, R. H. Y., & Fung, P. (2016) · Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d16-1110 · Cited by 126
“Thus the largest time contribution is due to the computations inside the network (He and Sun, 2015), which with a single convolution layer can be performed in negligible time for single utterances.”
BackgroundSignal Status Recognition Based on 1DCNN and Its Feature Extraction Mechanism Analysis
Huang, S., Tang, J., Dai, J., & Wang, Y. (2019) · Sensors · doi:10.3390/s19092018 · Cited by 137
“[35] explored the relationship between network depth, width and size of convolution kernel through a large number of experiments to optimize the network structure.”
BackgroundReal-time recognition system of soybean seed full-surface defects based on deep learning
Zhao, G., Quan, L., Li, H., Feng, H., Li, S., Zhang, S., & Liu, R. (2021) · Computers and Electronics in Agriculture · doi:10.1016/j.compag.2021.106230 · Cited by 145
Deep Learning-Based DOA Estimation
Zheng, S., Yang, Z., Shen, W., Zhang, L., Zhu, J., Zhao, Z., & Yang, X. (2024) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2024.3360527 · Cited by 160
“In the existing works [33], [34], [35], it has been pointed out that as the number of layers in a network increases, there is a likelihood of encountering issues such as gradient vanishing and gradient explosion.”
BackgroundOn Network Design Spaces for Visual Recognition
Radosavovic, I., Johnson, J., Xie, S., Lo, W.-Y., & Dollar, P. (2019) · 2019 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00197 · Cited by 127
“From prior work we know that a model’s error is related to its complexity; in particular more complex models are often more accurate [7, 34, 11, 39].”
BackgroundMethodKılıçarslan, S., Çelik, M., & Şahin, Ş. (2020) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2020.102231 · Cited by 142
Driver Drowsiness Detection Using Condition-Adaptive Representation Learning Framework
Yu, J., Park, S., Lee, S., & Jeon, M. (2018) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2018.2883823 · Cited by 146
“We have estimated the computational complexity of the proposed framework based on the approaches of He and Sun [57] and Notchenko et al.”
MethodA Reinforcement Learning and Blockchain-Based Trust Mechanism for Edge Networks
Xiao, L., Ding, Y., Jiang, D., Huang, J., Wang, D., Li, J., & Vincent Poor, H. (2020) · IEEE Transactions on Communications · doi:10.1109/tcomm.2020.2995371 · Cited by 123
“(1) in [35], the computational complexity of Algorithm 2 is given by”
BackgroundMethodMachine Learning-Based 5G-and-Beyond Channel Estimation for MIMO-OFDM Communication Systems
Le, H. A., Chien, T. V., Nguyễn, T. H., Choo, H., & Nguyễn, V. Đ. (2021) · Sensors · doi:10.3390/s21144861 · Cited by 146
“Therefore, the complexity of all convolution layers is O(∑l∈L k(2)l a 1 l a 2 l cl−1cl)[43].”
BackgroundLi, W., Guo, Y., Wang, B., & Yang, B. (2022) · Pattern Recognition · doi:10.1016/j.patcog.2022.109234 · Cited by 133
A deep learning solution approach for high-dimensional random differential equations
Nabian, M. A., & Meidani, H. (2019) · Probabilistic Engineering Mechanics · doi:10.1016/j.probengmech.2019.05.001 · Cited by 164
“However, deeper neural networks are more cumbersome to train, mainly due to the degradation problem: it is observed numerically, that with an increase in depth of neural networks, training and test accuracies get saturated and then degrade rapidly, possibly due to optimization issues [34, 35].”
BackgroundEfficient Short-Term Electricity Load Forecasting for Effective Energy Management
Khan, Z. A., Ullah, A., Haq, I. U., Hamdy, M., Mauro, G. M., Muhammad, K., Hijji, M., & Baik, S. W. (2022) · Sustainable Energy Technologies and Assessments · doi:10.1016/j.seta.2022.102337 · Cited by 161
Spatiotemporal imputation of MAIAC AOD using deep learning with downscaling
Li, L., Franklin, M., Girguis, M. S., Lurmann, F., Wu, J. Y., Pavlovic, N., Breton, C. V., Gilliland, F. D., & Habre, R. (2019) · Remote Sensing of Environment · doi:10.1016/j.rse.2019.111584 · Cited by 131
“To remedy this issue, residual connections of identity mapping can be introduced into the neural network to improve the learning efficiency, as shown in many applications of convolutional neural network (CNN) in deep learning (He and Sun 2015; He et al. 2016a, b).”
MethodMinimalistic CNN-based ensemble model for gender prediction from face images
Antipov, G., Berrani, S.-A., & Dugelay, J. (2015) · Pattern Recognition Letters · doi:10.1016/j.patrec.2015.11.011 · Cited by 120
“Specifically, 90% of its weights is taken up by the fully-connected layers and more than 90% of its running time is taken by the convolutional layers (He and Sun (2014)).”
BackgroundDomain Knowledge-Based Deep-Broad Learning Framework for Fault Diagnosis
Feng, J. Y., Yao, Y., Lu, S., & Liu, Y. (2020) · IEEE Transactions on Industrial Electronics · doi:10.1109/tie.2020.2982085 · Cited by 125
“Finally, our strategy takes more computational loads than the compared CNN in Table II (cid:8) [41].”
MethodRaiaan, M. A. K., Fatema, K., Khan, I. U., Azam, S., Rashid, M. R. U., Mukta, M. S. H., Jonkman, M., & Boer, F. D. (2023) · IEEE Access · doi:10.1109/access.2023.3272228 · Cited by 138
“We compute the theoretical time complexity, which is defined in [73] as follows : In the (15), the index of the convolutional layer is denoted by j and the number of convolutional layers is denoted by k . n j − 1 represents the number of filters (input channels) in the j - 1th layer, whereas n j…”
BackgroundMethodDeep Learning in Visual Computing and Signal Processing
Xie, D., Zhang, L., & Li, B. (2017) · Applied Computational Intelligence and Soft Computing · doi:10.1155/2017/1320780 · Cited by 136
“He and Sun [102] investigated training CNN under time cost constrains and proposed fast training methods for real-world applications while having similar performance as existing CNN models.”
MethodDeep Residual Networks with Exponential Linear Unit
Shah, A., Kadam, E., Shah, H., Shinde, S., & Shingade, S. (2016) · Proceedings of the Third International Symposium on Computer Vision and the Internet · doi:10.1145/2983402.2983406 · Cited by 112
“The training error increases as we add more layers to a deep model, as mentioned in [2].”
BackgroundDNN Transfer Learning From Diversified Micro-Doppler for Motion Classification
Seyfioğlu, M. S., Erol, B., Gürbüz, S. Z., & Amin, M. G. (2018) · IEEE Transactions on Aerospace and Electronic Systems · doi:10.1109/taes.2018.2883847 · Cited by 126
“In CNNs, depth has been shown to be more significant than width or filter size [58].”
BackgroundSmall-Object Detection in UAV-Captured Images via Multi-Branch Parallel Feature Pyramid Networks
Liu, Y., Yang, F., & Peng, H. (2020) · IEEE Access · doi:10.1109/access.2020.3014910 · Cited by 129
Kılıçarslan, S., Adem, K., & Çelik, M. (2020) · Medical Hypotheses · doi:10.1016/j.mehy.2020.109577 · Cited by 125
Nearest cluster-based intrusion detection through convolutional neural networks
Andresini, G., Appice, A., & Malerba, D. (2021) · Knowledge-Based Systems · doi:10.1016/j.knosys.2021.106798 · Cited by 121
Li, X., Xie, L., Wang, C., Miao, J., Shen, H., & Zhang, L. (2024) · GIScience & Remote Sensing · doi:10.1080/15481603.2024.2356355 · Cited by 130
“The last kind of metric is model complexity, which assesses the number of trainable parameters and the number of floating-point operations (FLOPs) (He and Sun, 2015; Molchanov et al. 2016), respectively.”
Márquez, E. Q. M., Hare, J. S., & Niranjan, M. (2018) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2018.2805098 · Cited by 149
“s and m is the size of the input and output (spatial size), respectively1 [29].”
BackgroundWang, J., Bai, Y., & Xia, B. (2020) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2020.3012547 · Cited by 123
“the design of deep neural networks [32].”
Li, X., Jiang, H., Liu, Y., Wang, T., & Li, Z. (2021) · Knowledge-Based Systems · doi:10.1016/j.knosys.2021.107652 · Cited by 115
Du, J., He, Z., Wang, L., Gholipour, A., Zhou, Z., Chen, D., & Jia, Y. (2019) · Neurocomputing · doi:10.1016/j.neucom.2018.10.102 · Cited by 129
Automated identification of retinopathy of prematurity by image-based deep learning
Tong, Y., Lü, W., Deng, Q., Chen, C., & Shen, Y. (2020) · Eye and Vision · doi:10.1186/s40662-020-00206-2 · Cited by 118
“However, as the network gains greater depth, its performance becomes saturated and then begins to decrease rapidly [26].”
BackgroundLandslide Detection Using Densely Connected Convolutional Networks and Environmental Conditions
Cai, H., Chen, T., Niu, R., & Plaza, A. (2021) · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · doi:10.1109/jstars.2021.3079196 · Cited by 125
“Thisconnection methodcanavoidtheproblemofgradientvanishingwiththe deepeningofthenetwork[48].”
VIPLFaceNet: an open source deep face recognition SDK
Liu, X., Kan, M., Wu, W., Shan, S., & Chen, X. (2017) · Frontiers of Computer Science · doi:10.1007/s11704-016-6076-3 · Cited by 118
Anti-Intelligent UAV Jamming Strategy via Deep Q-Networks
Gao, N., Qin, Z., Jing, X., Ni, Q., & Jin, S. (2019) · IEEE Transactions on Communications · doi:10.1109/tcomm.2019.2947918 · Cited by 129
“…Analysis: The total time complexity of anti-intelligent UAV jamming strategy mainly depends on the all convolutional layers, which can be defined as [39] O where m is the index of convolutional layer, the symbol F m − 1 is the number input channels of the m -th layer, i.e., F 0 = 1 , the symbol µ m…”
BackgroundA novel background subtraction algorithm based on parallel vision and Bayesian GANs
Zheng, W., Wang, K., & Wang, F. (2019) · Neurocomputing · doi:10.1016/j.neucom.2019.04.088 · Cited by 113
CPS data streams analytics based on machine learning for Cloud and Fog Computing: A survey
Fei, X., Shah, N., Verba, N., Chao, K., Sánchez-Anguix, V., Lewandowski, J., James, A. E., & Usman, Z. (2018) · Future Generation Computer Systems · doi:10.1016/j.future.2018.06.042 · Cited by 125
“We share the same view with the authors [131] that introducing computational time and memory constraints can give better understanding the value of machine learning methods in realistic business applications.”
BackgroundInfluentialPolyp-Net: A Multimodel Fusion Network for Polyp Segmentation
Banik, D., Roy, K., Bhattacharjee, D., Nasipuri, M., & Krejcar, O. (2020) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2020.3015607 · Cited by 109
“However, the pooling layers and the fully connected layers only consume ( 5–10 ) % of the computational time [37], whereas the convolutional layers consume the majority of the time.”
BackgroundAsphalt Pavement Crack Detection Based on Convolutional Neural Network and Infrared Thermography
Liu, F., Liu, J., & Wang, L. (2022) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2022.3142393 · Cited by 123
“The model and computational complexity also play an important role in evaluating CNN [12], [13].”
BackgroundReinforcement Learning-Based Microgrid Energy Trading With a Reduced Power Plant Schedule
Lu, X., Xiao, X., Xiao, L., Dai, C., Peng, M., & Poor, H. V. (2019) · IEEE Internet of Things Journal · doi:10.1109/jiot.2019.2941498 · Cited by 120
Intelligent Fault Diagnosis for Chemical Processes Using Deep Learning Multimodel Fusion
Wang, N., Yang, F., Zhang, R., & Gao, F. (2020) · IEEE Transactions on Cybernetics · doi:10.1109/tcyb.2020.3038832 · Cited by 108
Channel Estimation Using CNN-LSTM in RIS-NOMA Assisted 6G Network
Nguyen, C. M. H., Hoang, T. M., & Cheema, A. A. (2023) · IEEE Transactions on Machine Learning in Communications and Networking · doi:10.1109/tmlcn.2023.3278232 · Cited by 123
“…l is the index of a convolutional layer, d is the number of convolutional layers, n l is the number of filters or the width in l th layer, s w and s h are the width and the height of the filter respectively, and o w and o h are the width and height of the output feature map correspondingly [54].”
BackgroundLiu, L., Chen, S., Zhang, F., Wu, F., Pan, Y. F., & Wang, J. (2019) · Neural Computing and Applications · doi:10.1007/s00521-019-04096-x · Cited by 112
“As reported in the study of He and Sun, when deep networks start converging, the degradation problem has been exposed [42, 43].”
ResultHigh Dimensional Channel Estimation Using Deep Generative Networks
Balevi, E., Doshi, A. S., Jalal, A., Dimakis, A. G., & Andrews, J. G. (2020) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2020.3036947 · Cited by 138
Tang, Z., Tian, E., Wang, Y., Wang, L., & Yang, T. (2020) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2020.2985159 · Cited by 110
“In a CNN, the convolutional operation can extract local pattern in a transitionally invariant manner, and the computation complexity of the convolutional layer [22] is O(nl−1 · k2 l · nl ·m(2)l ), where nl−1 and nl are the output channel number of feature maps in the (l − 1)th layer and lth layer, respectively, kl is…”
Wei, H., Zhang, Q., Shang, M., & Gu, Y. (2021) · Measurement · doi:10.1016/j.measurement.2021.109864 · Cited by 111
Yu, Y., Karimi, H. R., Shi, P., Peng, R., & Zhao, S. (2024) · Mechanical Systems and Signal Processing · doi:10.1016/j.ymssp.2024.111194 · Cited by 109
Shallow convolutional neural network for image classification
Lei, F., Liu, X., Dai, Q., & Ling, B. W. (2019) · SN Applied Sciences · doi:10.1007/s42452-019-1903-4 · Cited by 117
“According to [23], we calculate the theoretical time complexity which is defined as: where j denotes the index of the convolutional layer, and k is the number of the convolutional layers. n j − 1 denotes the number of the filters (input channels) in the j − 1 th layer, and n j is the number of the…”
BackgroundReinforcement Learning Based PHY Authentication for VANETs
Lu, X., Xiao, L., Xu, T., Zhao, Y., Tang, Y., & Zhuang, W. (2020) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2020.2967026 · Cited by 105
“According to the CNN architecture in Algorithm 2, by [25], we have Thus, similar to [40] and [41], as s 1 = s 2 = 1 and n 1 > n 2 > n 3 , the computational complexity of the deep RL based PHY authentication scheme is given by O kZ”
MethodHyperspectral Band Selection Using Attention-Based Convolutional Neural Networks
Lorenzo, P. R., Tulczyjew, Ł., Marcinkiewicz, M., & Nalepa, J. (2020) · IEEE Access · doi:10.1109/access.2020.2977454 · Cited by 116
“In general, the time complexity of a 1D convolutional layer amounts to O(k · b · n), assuming a single input channel to this layer, and k denotes the kernel size, b is the size of the input tensor, and n is the number of kernels in this layer [87], [88].”
BackgroundZou, L., Li, Y., & Xu, F. (2020) · Neurocomputing · doi:10.1016/j.neucom.2020.04.074 · Cited by 104
An ANN Model for Predicting the Compressive Strength of Concrete
Lin, C., & Wu, N. (2021) · Applied Sciences · doi:10.3390/app11093798 · Cited by 105
“…sand, etc. Recently, more and more new types of ANN structures, such as the Support Vector Machines [17,18], Deep Learning Neural Network [19–23], Radial Basis Function Network [24–29] for various research areas have popped out, as well as other algorithms like tabu and genetic…”
Abnormal Event Detection and Localization via Adversarial Event Prediction
Yu, J., Lee, Y., Yow, K., Jeon, M., & Pedrycz, W. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3053563 · Cited by 116
Driver drowsiness detection using hybrid convolutional neural network and long short-term memory
Guo, J.-M., & Markoni, H. (2018) · Multimedia Tools and Applications · doi:10.1007/s11042-018-6378-6 · Cited by 113
“So the architecture has more capability to find a complex pattern, but replacing this filter into two different filters will increasing the depth and it will be not always giving better accuracy [5, 6].”
BackgroundOn defending against label flipping attacks on malware detection systems
Taheri, R., Javidan, R., Shojafar, M., Pooranian, Z., Miri, A., & Conti, M. (2020) · Neural Computing and Applications · doi:10.1007/s00521-020-04831-9 · Cited by 127
“method, which has a complexity of O(n). Similarly, in lines 6- 8, the model is based on the LP algorithm, which has a complexity of O(n). Then, lines 9 and 10 present CNN model creating, according to [15], which has a computational complexity of all convolutional layers. CNN computational complexity is O( d i=1n (l 1)…”
MethodRobust real-time emotion detection system using CNN architecture
Jaiswal, S., & Nandi, G. C. (2019) · Neural Computing and Applications · doi:10.1007/s00521-019-04564-4 · Cited by 107
“As discussed in [34], authors have compared time complexity of different variants of convolutional neural network.”
BackgroundZeng, X., Cao, K., & Zhang, M. (2017) · Proceedings of the 15th Annual International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3081333.3081336 · Cited by 103
“It is worthwhile to note that many similar strategies have been validated by other researchers in building compressed models for other tasks such as object recognition [13, 16].”
BackgroundMethodDeepEye: Deep convolutional network for pupil detection in real environments
Vera-Olmos, F. J., Pardo, E., Melero, H., & Malpica, N. (2018) · Integrated Computer-Aided Engineering · doi:10.3233/ica-180584 · Cited by 105
“Nonetheless, when a certain point of depth is reached, the performance of the model starts to degrade rapidly [43,44]; [25] shows that this degradation happens due to the difficulties of the solver to ap-F proximate identity maps by multiple non-linear layers.”
Wideband Channel Estimation With a Generative Adversarial Network
Balevi, E., & Andrews, J. G. (2021) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2020.3047100 · Cited by 106
“To prevent the training complexity of CNNs for channel estimation, [19], [20] proposed an untrained neural network that can precede or follow a least-squares (LS) estimator for OFDM and MIMOOFDM channels, respectively.”
InfluentialSingle Infrared Image Stripe Noise Removal Using Deep Convolutional Networks
Kuang, X., Sui, X., Chen, Q., & Gu, G. (2017) · IEEE photonics journal · doi:10.1109/jphot.2017.2717948 · Cited by 103
“4 Number of Layers: Research in [29] shows that adding more layers moderately could improve CNN.”
BackgroundDyNet: Dynamic Convolution for Accelerating Convolutional Neural NetworksPreprint
Zhang, Y., Zhang, J., Wang, Q., & Zhong, Z. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.10694 · Cited by 135
“The increasing demands of high-quality networks on mobile/embedding devices have driven the study on efficient network design (He & Sun, 2015).”
BackgroundA deep learning reconstruction framework for X-ray computed tomography with incomplete data
Dong, J., Fu, J., & Zhao, H. (2019) · PLoS ONE · doi:10.1371/journal.pone.0224426 · Cited by 107
“However, with the network’s depth increasing, accuracy gets saturated and then degrades rapidly as reported in [25, 26].”
BackgroundDeep Learning for SVD and Hybrid Beamforming
Peken, T., Adiga, S., Tandon, R., & Bose, T. (2020) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2020.3004386 · Cited by 107
Time Complexity in Deep Learning Models
Shah, B., & Bhavsar, H. (2022) · Procedia Computer Science · doi:10.1016/j.procs.2022.12.023 · Cited by 112
Deep transfer learning for military object recognition under small training set condition
Yang, Z. Q., Yu, W., Liang, P., Guo, H., Xia, L., Zhang, F., Ma, Y., & Ma, J. (2018) · Neural Computing and Applications · doi:10.1007/s00521-018-3468-3 · Cited by 103
“On the one hand, it has been pointed out that deep network structure plays the most important role in improving network’s performance [10].”
BackgroundEnsemble Learners of Multiple Deep CNNs for Pulmonary Nodules Classification Using CT Images
Zhang, B., Qi, S., Monkam, P., Li, C., Yang, F., Yao, Y., & Qian, W. (2019) · IEEE Access · doi:10.1109/access.2019.2933670 · Cited by 106
“time complexity than our single 2D CNN [43].”
BackgroundReinforcement Learning-Based Downlink Interference Control for Ultra-Dense Small Cells
Xiao, L., Zhang, H., Xiao, Y., Wan, X., Liu, S., Wang, L., & Poor, H. V. (2019) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2019.2945951 · Cited by 98
“Similar to the analysis in [39], the computational complexity of the DRLIC scheme in Algorithm 2 is O ( (cid:17) ) , in which f l − 1 denotes the number of input channels of the l -th Conv layer, and m l denotes the spatial size of the output feature map in the l -th Conv layer.”
BackgroundMethodA Survey of Online Data-Driven Proactive 5G Network Optimisation Using Machine Learning
Ma, B., Guo, W., & Zhang, J. (2020) · IEEE Access · doi:10.1109/access.2020.2975004 · Cited by 118
A context-sensitive deep learning approach for microcalcification detection in mammograms
Wang, J., & Yang, Y. (2018) · Pattern Recognition · doi:10.1016/j.patcog.2018.01.009 · Cited by 103
“Such choice was based on the observation that the number of Conv layers tends to have a larger impact on performance than other parameters (such as the number of filters and filter size) in the design of deep architectures [40].”
BackgroundAn Incentive Mechanism for Privacy-Preserving Crowdsensing via Deep Reinforcement Learning
Liu, Y., Wang, H., Peng, M., Guan, J., & Wang, Y. (2020) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.3047105 · Cited by 100
“According to [38], the time complexity for a Conv layer is (cid:16) , where L is the number of Conv layers, F is the length of feature map, K denotes the size of kernel, and C is the number of filters.”
BackgroundLightweight Automatic Modulation Classification Based on Decentralized Learning
Fu, X. G., Gui, G., Wang, Y., Ohtsuki, T. O., Adebisi, B., Gacanin, H., & Adachi, F. (2021) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2021.3089178 · Cited by 99
Predicting Household Electric Power Consumption Using Multi-step Time Series with Convolutional LSTM
Cascone, L., Sadiq, S., Ullah, S., Mirjalili, S., Siddiqui, H. U. R., & Umer, M. (2022) · Big Data Research · doi:10.1016/j.bdr.2022.100360 · Cited by 99
ME R-CNN: Multi-Expert R-CNN for Object Detection
Lee, H., Eum, S., & Kwon, H. (2019) · IEEE Transactions on Image Processing · doi:10.1109/tip.2019.2938879 · Cited by 106
“Several recently introduced measures such as multi-GPU parallel computing or enlarging mini-batch size [58], [59] can be taken into consideration to reduce the overall training time.”
BackgroundDeepOPF: Deep Neural Network for DC Optimal Power Flow
Pan, X., Zhao, T., & Chen, M. (2019) · IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm) · doi:10.1109/smartgridcomm.2019.8909795 · Cited by 136
“It can be evaluated by the scale of the network [21].”
Zhu, M., Li, Y., Pan, Z., & Yang, J. W. (2019) · Signal Processing · doi:10.1016/j.sigpro.2019.107393 · Cited by 91
An Improved Convolutional Neural Network for Three-Phase Inverter Fault Diagnosis
Zhang, S., Wang, R., Si, Y., & Wang, L. (2021) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2021.3129198 · Cited by 101
Kazemzadeh, M., Hisey, C. L., Zargar‐Shoshtari, K., Xu, W., & Broderick, N. G. R. (2022) · Optics Communications · doi:10.1016/j.optcom.2022.127977 · Cited by 95
Tomato Leaf Disease Detection System Based on FC-SNDPN
Huang, X., Chen, A., Zhou, G., Zhang, X., Wang, J., Peng, N. S., Yan, N., & Jiang, C. (2022) · Multimedia Tools and Applications · doi:10.1007/s11042-021-11790-3 · Cited by 101
“Meanwhile, ResNet was proposed in 2015 and won the first place in the classification task of ImageNet competition, because it was both “simple and practical” [18].”
BackgroundMa, S., Liu, W., Cai, W., Shang, Z., & Liu, G. (2019) · IEEE Access · doi:10.1109/access.2019.2912072 · Cited by 96
“Moreover, as the network depth increases, the accuracy of the model decreases [41].”
BackgroundVoltage Regulation of DC-DC Buck Converters Feeding CPLs via Deep Reinforcement Learning
Cui, C., Yan, N., Huangfu, B., Yang, T., & Zhang, C. (2021) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2021.3107535 · Cited by 94
“Remark 1: The time complexity of the proposed algorithm relies on the size of the DNN input n l , the number of neurons in the hidden layers, and the dimension of the DNN output m l [19].”
Practical Network Blocks Design with Q-Learning.Preprint
Zhong, Z., Yan, J., & Liu, C. (2017) · arXiv (Cornell University) · Cited by 99
Pasupa, K., Vatathanavaro, S., & Tungjitnob, S. (2020) · Journal of Ambient Intelligence and Humanized Computing · doi:10.1007/s12652-020-01773-x · Cited by 96
“1 ResNet It has been known that when a plain network is deep, the back-propagation gradient is small or vanishing, resulting in higher training and test errors (He et al., 2016; Huang et al., 2017; Srivastava et al., 2015; Glorot and Bengio, 2010; He and Sun, 2015).”
DeepFLASH: An Efficient Network for Learning-Based Medical Image Registration
Wang, J., & Zhang, M. (2020) · 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/cvpr42600.2020.00450 · Cited by 112
The Impact of Filter Size and Number of Filters on Classification Accuracy in CNN
Ahmed, W. S., & Karim, A. A. A. (2020) · doi:10.1109/csase48920.2020.9142089 · Cited by 89
“The present architecture in the ImageNet dataset achieves highly competitive accuracy (11.8% top-5 error, 10-view test), yet is 20% faster than “AlexNet” [ 8] (16.0% top-5 error, 10-view test) [9].”
InfluentialXiong, Y., & Zuo, R. (2021) · Mathematical Geosciences · doi:10.1007/s11004-021-09935-z · Cited by 92
Development of residual learning in deep neural networks for computer vision: A survey
Xu, G., Wang, X., Wu, X., Leng, X., & Xu, Y. (2024) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2024.109890 · Cited by 114
“Accuracy may reach a saturation point and then rapidly decrease [47] [48].”
Distribution-Oriented Aesthetics Assessment With Semantic-Aware Hybrid Network
Cui, C., Liu, H., Lian, T., Nie, L., Zhu, L., & Yin, Y. (2018) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2018.2875357 · Cited by 87
“In literature [43], the number of filters is termed the width of layers.”
BackgroundDeepTrend 2.0: A light-weighted multi-scale traffic prediction model using detrending
Dai, X., Fu, R., Zhao, E., Zhang, Z., Lin, Y., Wang, F., & Li, L. (2019) · Transportation Research Part C Emerging Technologies · doi:10.1016/j.trc.2019.03.022 · Cited by 88
Improving deep neural network with Multiple Parametric Exponential Linear Units
Li, Y., Fan, C., Li, Y., Wu, Q., & Ming, Y. (2018) · Neurocomputing · doi:10.1016/j.neucom.2018.01.084 · Cited by 98
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.