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 301–400 of 1,001
Abid, F., Li, C., & Alam, M. (2020) · Computer Communications · doi:10.1016/j.comcom.2020.04.002 · Cited by 63
Channel Non-Line-of-Sight Identification Based on Convolutional Neural Networks
Zheng, Q., He, R., Ai, B., Huang, C., Chen, W. R., Zhong, Z. D., & Zhang, H. (2020) · IEEE Wireless Communications Letters · doi:10.1109/lwc.2020.2994945 · Cited by 58
“Meanwhile, the training times of different networks are also tested [22].”
BackgroundPei, Z., Jiang, H., Li, X., Zhang, J., & Liu, S. (2021) · Measurement Science and Technology · doi:10.1088/1361-6501/abe5e3 · Cited by 58
“According to literature [45], the spatial complexity limits the minimum amount of data required for training, meanwhile the temporal complexity matters the optimization and validation of the model.”
Wang, Z., Ye, K., Jiang, M., Yao, J., Xiong, N. N., & Yen, G. G. (2021) · Swarm and Evolutionary Computation · doi:10.1016/j.swevo.2021.100975 · Cited by 58
Variance-based global sensitivity analysis for rear-end crash investigation using deep learning
Moussa, G., Owais, M., & Dabbour, E. (2021) · Accident Analysis & Prevention · doi:10.1016/j.aap.2021.106514 · Cited by 62
Deep Learning Techniques for the Classification of Brain Tumor: A Comprehensive Survey
Younis, A., Li, Q., Khalid, M., Clemence, B., & Adamu, M. J. (2023) · IEEE Access · doi:10.1109/access.2023.3317796 · Cited by 63
Owais, M., Alshehri, A. M., Gyani, J., Aljarbou, M. H., & Alsulamy, S. H. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.123114 · Cited by 61
Learning a Wavelet-Like Auto-Encoder to Accelerate Deep Neural Networks
Chen, T., Lin, L., Zuo, W., Luo, X., & Zhang, L. (2018) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v32i1.12282 · Cited by 64
Semi-supervised feature learning for improving writer identification
Chen, S., Wang, Y., Lin, C., Ding, W., & Cao, Z. (2019) · Information Sciences · doi:10.1016/j.ins.2019.01.024 · Cited by 64
Skin Disease Diagnosis from Photographs Using Deep Learning
Göçeri, E. (2019) · Lecture notes in computational vision and biomechanics · doi:10.1007/978-3-030-32040-9_25 · Cited by 62
Lightweight Modules for Efficient Deep Learning Based Image Restoration
Lahiri, A., Bairagya, S., Bera, S., Haldar, S., & Biswas, P. K. (2020) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2020.3007723 · Cited by 64
“Executing these large models on resource constrained platforms requires efficient design strategies [24].”
Deep CNN for Wideband Mmwave Massive Mimo Channel Estimation Using Frequency Correlation
Dong, P., Zhang, H., Li, G. Y., Naderializadeh, N., & Gaspar, I. S. (2019) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2019.8682819 · Cited by 62
“According to [14], the computational complexity of CNN processing is CCNN,2 ∼ O (∑Lc l=1M1,lM2,lF 2 l Nl−1Nl ) , where Lc is the number of convolutional layers, M1,l and M2,l denote the numbers 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,…”
MethodSky Imager-Based Forecast of Solar Irradiance Using Machine Learning
Al-lahham, A., Theeb, O., Elalem, K., Alshawi, T. A., & Alshebeili, S. A. (2020) · Electronics · doi:10.3390/electronics9101700 · Cited by 62
“The pooling and fully connected layers (FCL) take about 10% of the computational time [29].”
BackgroundVariational Self-attention Network for Sequential Recommendation
Zhao, J., Zhao, P., Zhao, L., Liu, Y., Sheng, V. S., & Zhou, X. (2021) · IEEE International Conference on Data Engineering (ICDE) · doi:10.1109/icde51399.2021.00138 · Cited by 70
“And adding more layers to a model of appropriate depth will lead to higher training errors, as reported in [27].”
BackgroundNalajam, P. K., & Varadarajan, R. (2021) · IEEE Access · doi:10.1109/access.2021.3097177 · Cited by 57
“CNN layer is estimated as O ( (cid:80) dl = 1 n l − 1 . s 2 l . n l . q 2 l ), where d , n l − 1 , s l and q l are the number of convolutional layers, number of filters in l th layer, spatial size of filter and spatial size of output feature map respectively [48].”
BackgroundDeep Residual Network for Smartwatch-Based User Identification through Complex Hand Movements
Mekruksavanich, S., & Jitpattanakul, A. (2022) · Sensors · doi:10.3390/s22083094 · Cited by 62
“While enhancing the level of a CNN could extract more abstract features and improve effectiveness, it may also result in degraded performance [23].”
BackgroundAn Overview of Methods for Deep Learning in Neural Networks
Созыкин, А. В. (2017) · Bulletin of the South Ural State University Series Computational Mathematics and Software Engineering · doi:10.14529/cmse170303 · Cited by 61
Guo, R., Liu, J., Li, N., Liu, S., Fu, C., Cheng, B., Duan, J., Li, X., & Ma, C. (2018) · ISPRS International Journal of Geo-Information · doi:10.3390/ijgi7030110 · Cited by 64
“The total time complexity [61] of convolutional layers is Thus, computational complexity becomes a vital factor to consider.”
BackgroundXue, J., Xiang, Z., & Ou, G. (2021) · Engineering Structures · doi:10.1016/j.engstruct.2021.111859 · Cited by 61
A Deep Learning Framework for Robust DOA Estimation Using Spherical Harmonic Decomposition
Varanasi, V., Gupta, H., & Hegde, R. M. (2020) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2020.2984852 · Cited by 64
“…complexity for SH-MUSIC-DPD is therefore 2) Proposed SH-P-PM-CNN Approach: The run-time complexity of the CNN based DOA estimation is given by where the first and second summation represents the run time complexity for the convolutional (conv) [64], [65] and FC layers [66]–[68], respectively.”
Adapting to Dynamic LEO-B5G Systems: Meta-Critic Learning Based Efficient Resource Scheduling
Yuan, Y., Lei, L., Vu, T. X., Chang, Z., Chatzinotas, S., & Sun, S. (2022) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2022.3178171 · Cited by 63
“The time complexity of CNN is O (∑V1 v=1 oc,v−1%v ) , where %v = o(2)k,voc,vo 2 f,v [41].”
BackgroundYin, G., Sun, S., Yu, D., Li, D., & Zhang, K. (2022) · ACM Transactions on Multimedia Computing Communications and Applications · doi:10.1145/3490686 · Cited by 61
ECG-Based Driver’s Stress Detection Using Deep Transfer Learning and Fuzzy Logic Approaches
Amin, M., Ullah, K., Asif, M., Waheed, A., Haq, S. U., Zareei, M., & Biswal, R. R. (2022) · IEEE Access · doi:10.1109/access.2022.3158658 · Cited by 58
“ResNet-101 is based on the residual connection, which skips some layers to reduce degradation [57].”
BackgroundMethodLow-dose COVID-19 CT Image Denoising Using CNN and its Method Noise Thresholding
Singh, P., Diwakar, M., Pandey, N. K., Singh, R., Sisodia, D., Arya, C., & Chakraborty, C. (2022) · Current Medical Imaging Formerly Current Medical Imaging Reviews · doi:10.2174/1573405618666220404162241 · Cited by 55
Rezaee, K., Khosravi, M. R., Attar, H. H., Menon, V. G., Khan, M. A., Issa, H., & Qi, L. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3284056 · Cited by 58
“In [18], it is proven that the temporal complexity of a CNN for a single sample may be described as for a CNN with d convolutional layers SpF and SpK are the spatial sizes of the output feature map and kernel, inMap i − 1 and inMap i are the number of input kernels and maps in the i th…”
BackgroundPelican: A Deep Residual Network for Network Intrusion Detection
Wu, P., Guo, H., & Moustafa, N. (2020) · International Conference on Dependable Systems and Networks workshops · doi:10.1109/dsn-w50199.2020.00018 · Cited by 68
FLOP-Reduction Through Memory Allocations Within CNN for Hyperspectral Image Classification
Paoletti, M. E., Haut, J. M., Tao, X., Plaza, J., & Plaza, A. (2020) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2020.3024730 · Cited by 58
A Hybrid Analysis-Based Approach to Android Malware Family Classification
Ding, C. Y., Luktarhan, N., Lu, B., & Zhang, W. (2021) · Entropy · doi:10.3390/e23081009 · Cited by 63
“However, this degradation is not caused by overfitting, and adding more layers to a suitably deep model leads to more training errors [28,29].”
BackgroundTwo-Stage Channel Estimation for mmWave Massive MIMO Systems Based on ResNet-UNet
Zhao, J., Wu, Y., Zhang, Q., & Liao, J. (2023) · IEEE Systems Journal · doi:10.1109/jsyst.2023.3234048 · Cited by 59
Recursive Spatial Transformer (ReST) for Alignment-Free Face Recognition
Wu, W., Kan, M., Liu, X., Yang, Y., Shan, S., & Chen, X. (2017) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.407 · Cited by 55
“1 is the theoretical time complexity [4] excluding the ReST compared to AlexNet and comp.”
A Highly Accurate Deep Learning Based Approach for Developing Wireless Sensor Network Middleware
Alshinina, R., & Elleithy, K. M. (2018) · IEEE Access · doi:10.1109/access.2018.2844255 · Cited by 59
“Literature has shown that while the training errors tend to decrease, errors increase with low accuracy after a while [47].”
BackgroundEnhancing the Fuel-Economy of V2I-Assisted Autonomous Driving: A Reinforcement Learning Approach
Liu, X., Liu, Y., Chen, Y., & Hanzo, L. (2020) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2020.2996187 · Cited by 61
“As demonstrated in [50], the total complexity of the CNN model is on the order of O( ∑2 i=1 fi−1fin 2 imi), where f0 is the number of input channels for the CNN model, and mi is the spatial size of the output feature map of the Conv layer i.”
BackgroundFedSCR: Structure-based Communication Reduction for Federated Learning
Wu, X., Yao, X., & Wang, C. (2020) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2020.3046250 · Cited by 70
“There are lots of studies attempting to understand the features of deep neural networks, including the function of intermediate feature layers [22], the impacts of network depth and width [22], [23] and different loss functions [24].”
BackgroundDeep Learning Strategy for Braille Character Recognition
Kausar, T., Manzoor, S., Kausar, A., Lu, Y. H., Muhammad, W., & Ashraf, M. A. (2021) · IEEE Access · doi:10.1109/access.2021.3138240 · Cited by 65
Zhou, Y., Chang, H., Lu, X., & Lu, Y. (2022) · Knowledge-Based Systems · doi:10.1016/j.knosys.2022.109658 · Cited by 59
Zhou, K. Q., Qin, Y., & Yuen, C. (2022) · IEEE/ASME Transactions on Mechatronics · doi:10.1109/tmech.2022.3201010 · Cited by 56
“has a training time complexity of per time step o(W ) [45], where W is the total number of weights in the network, and the model development is shown in Fig.”
Yang, C., Yang, C., Zhang, X., & Zhang, J. (2023) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2023.3248059 · Cited by 53
Fine-Tuning ResNet for Breast Cancer Classification from Mammography
Chen, Y., Zhang, Q., Wu, Y., Лю, Б., Wang, M., & Lin, Y. (2019) · Lecture notes in electrical engineering · doi:10.1007/978-981-13-6837-0_7 · Cited by 59
DeepCU: Integrating both Common and Unique Latent Information for Multimodal Sentiment Analysis
Verma, S., Wang, C., Zhu, L., & Liu, W. (2019) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2019/503 · Cited by 55
Wu, J., Xu, L., Wu, F., Kong, Y., Senhadji, L., & Shu, H. (2020) · Neurocomputing · doi:10.1016/j.neucom.2020.02.053 · Cited by 60
Attallah, O. (2023) · Chemosensors · doi:10.3390/chemosensors11070364 · Cited by 61
Deep Expander Networks: Efficient Deep Networks from Graph Theory
Prabhu, A., Varma, G., & Namboodiri, A. (2018) · Lecture notes in computer science · doi:10.1007/978-3-030-01261-8_2 · Cited by 71
Indoor Semantic Segmentation for Robot Navigating on Mobile
Kim, W., & Seok, J. (2018) · International Conference on Ubiquitous and Future Networks · doi:10.1109/icufn.2018.8436956 · Cited by 52
Vaddi, R., & Manoharan, P. (2020) · Infrared Physics & Technology · doi:10.1016/j.infrared.2020.103457 · Cited by 52
BayLIME: Bayesian Local Interpretable Model-Agnostic Explanations
Zhao, X., Huang, W., Huang, X., Robu, V., & Flynn, D. (2020) · Conference on Uncertainty in Artificial Intelligence · doi:10.48550/arxiv.2012.03058 · Cited by 154
“1 (B), to improve LIME’s efficiency, we cannot effectively tune other LIME arguments, rather the best option is to limit the number of queries made to the AI model (which, indeed, is very costly when the AI model is deep [He and Sun, 2015]) when generating labels for the n samples (step 3 of the LIME workflow in Appen…”
BackgroundPattern Recognition of Partial Discharge Image Based on One-dimensional Convolutional Neural Network
Wan, X., Song, H., Luo, L., Li, Z., Sheng, G., & Jiang, X. (2018) · 2018 Condition Monitoring and Diagnosis (CMD) · doi:10.1109/cmd.2018.8535761 · Cited by 51
“In the case of limited time complexity, the depth of the network, the number of feature channels and the size of the convolution filters all have an effect on the experimental results [9].”
BackgroundExpressivity of Deep Neural Networks
Gühring, I., Raslan, M., & Kutyniok, G. (2022) · Cambridge University Press eBooks · doi:10.1017/9781009025096.004 · Cited by 93
Vertex and energy reconstruction in JUNO with machine learning methods
Qian, Z., Belavin, V., Bokov, V., Brugnera, R., Compagnucci, A., Gavrikov, A., Garfagnini, A., Gonchar, M., Khatbullina, L., Li, Z., et al. (2021) · Nuclear Instruments and Methods in Physics Research Section A Accelerators Spectrometers Detectors and Associated Equipment · doi:10.1016/j.nima.2021.165527 · Cited by 68
“In order to solve this problem, we use ResNet network architecture [45].”
MethodRobust Single-Shot T2 Mapping via Multiple Overlapping-Echo Acquisition and Deep Neural Network
Zhang, J., Wu, J., Chen, S., Zhang, Z., Cai, S., Cai, C., & Chen, Z. (2019) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2019.2896085 · Cited by 52
“Such degradation could not be attributed to overfitting [38], [39].”
BackgroundDSLN: Securing Internet of Things Through RF Fingerprint Recognition in Low-SNR Settings
Wu, W., Hu, S., Lin, D., & Liu, Z. (2021) · IEEE Internet of Things Journal · doi:10.1109/jiot.2021.3100398 · Cited by 51
“The authors in [27] proved that the learning process of a neural network with a depth of d takes a time of O s 2 d , where s refers to the input dimension and O ( .”
BackgroundDong, B., Liu, Y., Gui, G., Fu, X. G., Dong, H., Adebisi, B., Gacanin, H., & Sari, H. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3194508 · Cited by 54
Electricity Theft Detection Based on Contrastive Learning and Non-Intrusive Load Monitoring
Gao, A., Mei, F., Zheng, J., Sha, H., Guo, M., & Xie, Y. (2023) · IEEE Transactions on Smart Grid · doi:10.1109/tsg.2023.3263219 · Cited by 50
Zhou, L., Li, Q., Huo, G., & Zhou, Y. (2017) · Computational Intelligence and Neuroscience · doi:10.1155/2017/3792805 · Cited by 52
“unchanged, a deeper structure does not always guarantee a better result, which has been validated by many experiments [16].”
ResultLook-behind fully convolutional neural network for computer-aided endoscopy
Diamantis, D. E., Iakovidis, D. K., & Koulaouzidis, A. (2018) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2018.12.005 · Cited by 52
A modular CNN-based building detector for remote sensing images
Konstantinidis, D., Argyriou, V., Stathaki, T., & Grammalidis, N. (2019) · Computer Networks · doi:10.1016/j.comnet.2019.107034 · Cited by 55
“Lately, 2D and 3D building detection from remote sensing images is tackled by means of machine learning and, more specifically, convolutional neural networks [7, 8, 9].”
BackgroundIntelligent Resource Allocation Schemes for UAV-Swarm-Based Cooperative Sensing
Li, T., Leng, S., Wang, Z., Zhang, K., & Zhou, L. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3183099 · Cited by 54
“Since the fully connected layers and pooling layers often consume 5%–10% of the computation time of a CNN [20], we consider only the computation of the convolutional layers, where L = { 1 , 2 , . . . , L } denotes the index set of the convolutional layers of the CNN.”
BackgroundCombining max-pooling and wavelet pooling strategies for semantic image segmentation
Brito, A. D. S., Vieira, M. B., de Andrade, M. L. S. C., Feitosa, R. Q., & Giraldi, G. A. (2021) · Expert Systems with Applications · doi:10.1016/j.eswa.2021.115403 · Cited by 49
Heating load prediction of residential district using hybrid model based on CNN
Zhao, A., Lu, M., Xue, X., Xi, J., & Jiao, Y. (2022) · Energy and Buildings · doi:10.1016/j.enbuild.2022.112122 · Cited by 48
Zhang, Z., Hwang, M., Kilbaugh, T. J., & Katz, J. (2023) · Measurement Science and Technology · doi:10.1088/1361-6501/ad1671 · Cited by 52
“Residual learning prevents the vanishing and/or exploding gradients [30] as well as the degradation problems [31] in deep learning.”
BackgroundAn Intelligent Fuzzing Data Generation Method Based on Deep Adversarial Learning
Li, Z., Zhao, H., Shi, J., Huang, Y., & Xiong, J. (2019) · IEEE Access · doi:10.1109/access.2019.2911121 · Cited by 53
“One of the purposes of this architecture design is to reduce the computational cost [50] of the model.”
BackgroundSNDCNN: Self-Normalizing Deep CNNs with Scaled Exponential Linear Units for Speech Recognition
Huang, Z., Ng, T., Liu, L., Mason, H., Zhuang, X., & Liu, D. (2020) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp40776.2020.9053973 · Cited by 57
“Accuracy often saturates and degrades as network depth increases [6, 7].”
BackgroundZhou, Q., Shi, Y. H., Xu, Z., Qu, R., & Xu, G. (2020) · IEEE Access · doi:10.1109/access.2020.2998098 · Cited by 48
“So, our SNNs are very friendly to smartphones and portable devices with low computational power (CPUs or low-end GPUs) limitations [82].”
BackgroundLearning Stacked Image Descriptor for Face Recognition
Lei, Z., Yi, D., & Li, S. Z. (2015) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2015.2473415 · Cited by 51
“As described in [14], the complexity of convolution operator is about O( ∑d l=1 nl−1 · s2 l nlm 2 l ), where l is the index of convolutional layer, d is the number of convolutional layers, nl is the number of learned filters at the lth layer, and nl−1 is the number of input channels of the lth layer.”
BackgroundA Transfer Double Deep Q Network Based DDoS Detection Method for Internet of Vehicles
Li, Z., Kong, Y., & Jiang, C. (2023) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2022.3233880 · Cited by 47
Deep Reinforcement Learning-Based Grant-Free NOMA Optimization for mURLLC
Yan, L., Deng, Y., Zhou, H. X., Elkashlan, M., & Nallanathan, A. (2023) · IEEE Transactions on Communications · doi:10.1109/tcomm.2023.3238061 · Cited by 55
“As demonstrated in [48], the computational complexity of the DRL algorithm can be calculated as O ( ¯ m ¯ n log¯ n ) , where ¯ m is the number of layers, and ¯ n is the number of units per learning layer.”
BackgroundZhang, Y., Wong, G., Mann, G. J., Müller, S., & Yang, J. Y. H. (2022) · GigaScience · doi:10.1093/gigascience/giac071 · Cited by 54
Photograph aesthetical evaluation and classification with deep convolutional neural networks
Tan, Y., Tang, P., Zhou, Y., Luo, W., Kang, Y., & Li, G. (2016) · Neurocomputing · doi:10.1016/j.neucom.2016.08.098 · Cited by 46
Towards Explainable Ear Recognition Systems Using Deep Residual Networks
Alshazly, H. A., Linse, C., Barth, E., Idris, S. A., & Martinetz, T. (2021) · IEEE Access · doi:10.1109/access.2021.3109441 · Cited by 55
“The second drawback is the performance degradation problem, as adding more layers to a network increases the parameter space that needs to be optimized and the training error can even increase [35], [36].”
BackgroundWang, B., Xu, K., Zheng, S., Zhou, H., & Liu, Y. (2022) · IEEE Transactions on Reliability · doi:10.1109/tr.2022.3148114 · Cited by 52
An integrated spatiotemporal-based methodology for deepfake detection
Ismail, A., Elpeltagy, M. S., Zaki, M. S., & ElDahshan, K. A. (2022) · Neural Computing and Applications · doi:10.1007/s00521-022-07633-3 · Cited by 53
“For simplicity, the time complexity of the proposed method is assessed based on the convolution layers [65–67].”
BackgroundMethodChen, Z., Ke, H., Xu, J., Peng, T., & Yang, C. (2022) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2022.3224988 · Cited by 47
“3) Model Complexity: As given in Table VI, the time complexity TCNN and space complexity SCNN of CNN are calculated by the convolution operation [24].”
MethodZhang, Y., Li, S., Li, D., Zhu, J., & Guan, Q. (2024) · IEEE Internet of Things Journal · doi:10.1109/jiot.2024.3372060 · Cited by 54
“According to [45], the computational complexity of a CNN module at each instant can be represented as O ( (cid:14) N L l = 1 a l b l s l k l ) .”
BackgroundMultiscale road centerlines extraction from high-resolution aerial imagery
Liu, R., Miao, Q., Song, J., Quan, Y., Li, Y., Xu, P., & Dai, J. (2018) · Neurocomputing · doi:10.1016/j.neucom.2018.10.036 · Cited by 48
Identifying critical nodes in temporal networks by network embedding
Yu, E., Fu, Y., Chen, X., Xie, M., & Chen, D. (2020) · Scientific Reports · doi:10.1038/s41598-020-69379-z · Cited by 53
Jia, G., Lam, H., & Althoefer, K. A. (2021) · Pattern Recognition · doi:10.1016/j.patcog.2021.108226 · Cited by 52
Zhang, J., Luo, Y., Wang, B., Lü, C., Si, J., & Song, J. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3121757 · Cited by 46
“As for CNN structure in ConvLSTM block, the temporal computational complexity of CNN structure per time step is (cid:25) [36], where L is the depth of Con-vLSTM structure, l denotes the l -th ConvLSTM layer, A is the spatial size of the output feature map of CNN, in this article, A = n , B l , C l…”
MethodAIA-Net: Adaptive Interactive Attention Network for Text–Audio Emotion Recognition
Zhang, T., Li, S., Chen, B., Yuan, H., & Chen, C. L. P. (2022) · IEEE Transactions on Cybernetics · doi:10.1109/tcyb.2022.3195739 · Cited by 52
“However, the vanishing gradients and degradation problem [61], [62] often occur during training when simply stacking deep learning layers.”
BackgroundQin, Y., Yuen, C., Yin, X., & Huang, B. (2022) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2022.3205942 · Cited by 48
“Second, for TemCap, the time complexity of the BCL is o(nb × s(2)b ×m(2)b) [25], where nb is the number of filters, sb is the spatial size of the filter, and mb is the spatial size of the output feature map.”
BackgroundEffective Quantization Approaches for Recurrent Neural Networks
Alom, M. Z., Moody, A. T., Maruyama, N., Essen, B. V., & Taha, T. M. (2018) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2018.8489341 · Cited by 58
Fast Single Image Super-Resolution Via Dilated Residual Networks
Zhang, L., Zhang, Y., Yali, P., Shigang, L., Wu, X., Lü, G., & Yuan, R. (2018) · IEEE Access · doi:10.1109/access.2018.2865613 · Cited by 48
“Meanwhile, as the layer rises, a deep network leads to higher training error, as reported in [14].”
BackgroundSupervised Deep Sparse Coding Networks for Image Classification
Sun, X., Nasrabadi, N. M., & Tran, T. D. (2019) · IEEE Transactions on Image Processing · doi:10.1109/tip.2019.2928121 · Cited by 54
“For instance, the imperative for sparse codes to adequately reconstruct an input signal [6] ameliorates information degeneracy issues within deep architectures [17], [21].”
BackgroundReal-Time Streaming Point Cloud Compression for 3D LiDAR Sensor Using U-Net
Tu, C., Takeuchi, E., Carballo, A., & Takeda, K. (2019) · IEEE Access · doi:10.1109/access.2019.2935253 · Cited by 56
Boggavarapu, L. N. P. K., & Manoharan, P. (2020) · Infrared Physics & Technology · doi:10.1016/j.infrared.2020.103455 · Cited by 44
Dual-Intended Deep Learning Model for Breast Cancer Diagnosis in Ultrasound Imaging
Vigil, N., Barry, M., Amini, A., Akhloufi, M. A., Maldague, X., Ma, L., Ren, L., & Yousefi, B. (2022) · Cancers · doi:10.3390/cancers14112663 · Cited by 46
“The major computational complexity of the convolutional neural networks concerns convolutional layers, which leads to 90–95% of the overall complexity of the model [58,59].”
BackgroundKe, H., Wang, F., Bi, H., Ma, H., Wang, G., & Yin, B. (2023) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.122853 · Cited by 52
SET: Spectral Enhancement for Tiny Object Detection
Sun, H., Wang, R., Li, Y., Yang, L., Lin, S., Cao, X., & Zhang, B. (2025) · 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/cvpr52734.2025.00444 · Cited by 70
“Recent advances in Deep Neural Networks (DNNs) [14, 60] have significantly improved the object detection field [11, 22].”
Wang, Y., Sun, L., & Jin, Q. (2019) · IEEE Transactions on Computational Biology and Bioinformatics · doi:10.1109/tcbb.2019.2911947 · Cited by 45
“The time complexity of the convolution layer[42] is shown in Equation (5), where l is the layer of convolution, d is the total number of the convolution layers, n l represents the number of channels of the feature diagram, n l (cid:4) 1 is also known as the number of input channels of the l th…”
BackgroundLi, Y., Xu, Y., Xu, Y., Liu, X., Wang, X., Li, W., & Anpalagan, A. (2020) · IEEE Wireless Communications Letters · doi:10.1109/lwc.2020.2999333 · Cited by 43
“The computation of the algorithms are calculated according to [9].”
MethodEEG signal classification based on SVM with improved squirrel search algorithm
Shi, M., Wang, C., Li, X., Li, M.-Q., Wang, L., & Xie, N. (2020) · Biomedizinische Technik/Biomedical Engineering · doi:10.1515/bmt-2020-0038 · Cited by 45
“(27) only includes the computational complexity of convolutional layer, excluding pooling layer and full connection layer [69].”
BackgroundPhotonic Convolution Neural Network Based on Interleaved Time-Wavelength Modulation
Jiang, Y., Zhang, W., Yang, F., & He, Z. (2021) · Journal of Lightwave Technology · doi:10.1109/jlt.2021.3076070 · Cited by 48
“In the CNN architecture, convolution layers will occupy most of computing time and resources [5] due to high computation complexity of multiply accumulate operation and matrices multiply accumulate operation (MMAC) [6].”
BackgroundHuang, W., Chang, W., Yan, G., Yang, Z., Luo, H., & Pei, H. (2021) · Expert Systems with Applications · doi:10.1016/j.eswa.2021.115968 · Cited by 50
Chung, W. J., & Liu, C. (2022) · Energy and Buildings · doi:10.1016/j.enbuild.2022.112521 · Cited by 50
Multiclass ECG Signal Analysis Using Global Average-Based 2-D Convolutional Neural Network Modeling
Wasimuddin, M., Elleithy, K. M., Abuzneid, A., Faezipour, M., & Abuzaghleh, O. (2021) · Electronics · doi:10.3390/electronics10020170 · Cited by 46
“The complexity of a CNN can be calculated in the form of Big O notations [81], expressed by Equation (6).”
BackgroundDeep Learning-Based Signal Detection for Underwater Acoustic OTFS Communication
Zhang, Y., Zhang, S., Wang, B., Liu, Y., Bai, W., & Shen, X. (2022) · Journal of Marine Science and Engineering · doi:10.3390/jmse10121920 · Cited by 44
“According to [36], the computational complexity of the CNN can be defined as O(∑l l=1 Nincl · w 2 · N fl (NM)), where Nl is the number of CNN layers used to construct the model, l is the index of a convolutional layer, N fl is the number of filters (also known as the width) in the lth layer, Nincl is the number of inp…”
A Lightweight Deep Learning-Based Model for Tomato Leaf Disease Classification
Ullah, N., Khan, J. A., Almakdi, S., Alshehri, M. S., Qathrady, M. A., Aldakheel, E. A., & Khafaga, D. S. (2023) · Computers, materials & continua/Computers, materials & continua (Print) · doi:10.32604/cmc.2023.041819 · Cited by 49
“Delayed downsampling (large activation maps) results in higher classification accuracies [34].”
A novel hybrid ensemble based Alzheimer’s identification system using deep learning technique
Ayus, I., & Gupta, D. K. (2024) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2024.106079 · Cited by 50
Exploiting artificial intelligence for digitally enriched museum visits
Ioannakis, G., Bampis, L., & Koutsoudis, A. (2019) · Journal of Cultural Heritage · doi:10.1016/j.culher.2019.07.019 · Cited by 46
George, K., Sankaran, P., & Paul Joseph, K. (2020) · Computer Methods and Programs in Biomedicine · doi:10.1016/j.cmpb.2020.105531 · Cited by 47
Lanjewar, M. G., Morajkar, P. P., & Parab, J. S. (2022) · Multimedia Tools and Applications · doi:10.1007/s11042-022-12392-3 · Cited by 42
Dewangan, D. K., Sahu, S. P., Sairam, B., & Agrawal, A. (2021) · Computing · doi:10.1007/s00607-021-00974-2 · Cited by 46
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.