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 701–800 of 1,001
Feature Evaluation of Deep Convolutional Neural Networks for Object Recognition and DetectionPreprint
Kataoka, H., Iwata, K., & Satoh, Y. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1509.07627 · Cited by 30
“[5], the most important CNN feature is deep architecture.”
BackgroundPham, H., Khoudour, L., Crouzil, A., Zegers, P., & Velastín, S. A. (2018) · Proceedings - International Conference on Image Processing · doi:10.1109/icip.2018.8451404 · Cited by 19
The State of the Art in Implementing Machine Learning for Mobile Apps: A Survey
Dai, X., Spasić, ., Chapman, S., & Meyer, B. (2020) · IEEE SoutheastCon · doi:10.1109/southeastcon44009.2020.9249652 · Cited by 21
Representation learning using Attention Network and CNN for Heterogeneous networks
Tong, N., Tang, Y., Chen, B., & Xiong, L. (2021) · Expert Systems with Applications · doi:10.1016/j.eswa.2021.115628 · Cited by 18
Global-local neighborhood based network representation for citation recommendation
Cai, X., Wang, N., Yang, L., & Mei, X. (2022) · Applied Intelligence · doi:10.1007/s10489-021-02964-5 · Cited by 18
“’s experiment [45] also proves that the deeper network performance will be relatively improved for two kinds of network structures with the same time complexity.”
BackgroundLimited Data Spectrum Sensing Based on Semi-Supervised Deep Neural Network
Zhang, Y., & Zhao, Z. (2021) · IEEE Access · doi:10.1109/access.2021.3135568 · Cited by 18
Diagnosis of exercise-induced cardiac fatigue based on deep learning and heart sounds
Yin, C., Zhou, X., Zhao, Y., Zheng, Y., Shi, Y. H., Yan, X., & Guo, X. (2022) · Applied Acoustics · doi:10.1016/j.apacoust.2022.108900 · Cited by 19
A whale optimization algorithm–based cellular automata model for urban expansion simulation
Ding, Y., Cao, K., Qiao, W., Shao, H., Yang, Y., & Li, H. (2022) · International Journal of Applied Earth Observation and Geoinformation · doi:10.1016/j.jag.2022.103093 · Cited by 18
Location and timestamp-based chip contour detection using LWMG-YOLOv5
Chang, B. R., Tsai, H. F., & Hsieh, C.-W. (2023) · Computers & Industrial Engineering · doi:10.1016/j.cie.2023.109277 · Cited by 18
Abid, F., Alam, M., Alamri, F. S., & Siddique, I. (2023) · AIMS Mathematics · doi:10.3934/math.20231019 · Cited by 23
Liu, D., Feng, G., Feng, G., & Xie, L. (2024) · SPE Journal · doi:10.2118/218418-pa · Cited by 19
Deep Learning-Assisted Signal Detection for OTFS-NOMA Systems
Umakoğlu, İ., Namdar, M., & Başgümüş, A. (2024) · IEEE Access · doi:10.1109/access.2024.3449812 · Cited by 19
“With the aid of the references, [25] and [35], the computational complexity of the BiLSTM can be expressed as O (( NM ) 2 ) where ( NM ) is the OTFS frame size for the network input.”
MethodWang, H., Hua, Z., Li, H., & Ju, X. (2024) · Computers, materials & continua/Computers, materials & continua (Print) · doi:10.32604/cmc.2024.058785 · Cited by 19
Park, K., Kim, J., & Lee, J. W. (2020) · Scientific Reports · doi:10.1038/s41598-020-62022-x · Cited by 20
Ameen, T., Chen, L., Xu, Z., Lyu, D., & Shi, H. (2020) · ISPRS International Journal of Geo-Information · doi:10.3390/ijgi9080464 · Cited by 18
PPDet: A novel infrared pedestrian detection network in a per-pixel prediction fashion
Wang, D., & Lan, J. (2021) · Infrared Physics & Technology · doi:10.1016/j.infrared.2021.103965 · Cited by 19
Wang, H., Liu, C., Du, W., & Wang, S. (2021) · Applied Sciences · doi:10.3390/app112311325 · Cited by 19
Feng, J., Jiang, L., Yan, L., Yi, A., Pan, W., & Luo, B. (2022) · Journal of Lightwave Technology · doi:10.1109/jlt.2022.3149412 · Cited by 18
Deep learning and its application to general image classification
Liu, P.-H., Su, S., Chen, M.-C., & Hsiao, C.-C. (2015) · International Conference on Informative and Cybernetics for Computational Social Systems · doi:10.1109/iccss.2015.7281139 · Cited by 18
ANTNets: Mobile Convolutional Neural Networks for Resource Efficient Image ClassificationPreprint
Xiong, Y., Kim, H. J., & Hedau, V. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1904.03775 · Cited by 24
“To this end, [4] studied module designs with the trade-off between multiple factors such as depth, width, filter size, pooling layer and so on.”
BackgroundCodnNet: A lightweight CNN architecture for detection of COVID-19 infection
Yang, J., Zhang, L., Tang, X., & Man, H. (2022) · Applied Soft Computing · doi:10.1016/j.asoc.2022.109656 · Cited by 21
“Model complexity includes time complexity and space complexity [44].”
BackgroundLi, W., Xu, Y., Chen, J., Yuan, H., Han, H., Xu, Y., & Feng, Z. (2022) · IEEE Wireless Communications Letters · doi:10.1109/lwc.2022.3219434 · Cited by 17
“92 × 10 7 FLOPs at the testing phase [14].”
BackgroundIndoor Semantic Location Privacy Protection With Safe Reinforcement Learning
Min, M., Yang, S. H., Zhang, H., Ding, J., Peng, G., Pan, M., & Han, Z. (2023) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2023.3291364 · Cited by 18
A law of data separation in deep learning
He, H., & Su, W. (2023) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2221704120 · Cited by 65
“S10 in SI Appendix shows that the law does not appear when batch normalization is not used.”
BackgroundAl-Barazanchi, H. A., Qassim, H., & Verma, A. (2016) · IEEE Annual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON) · doi:10.1109/uemcon.2016.7777858 · Cited by 17
“Degradation leads to higher training error as reported in [20, 21] when extending the network depth by adding more layers.”
BackgroundCharacter-Based Convolutional Grid Neural Network for Breast Cancer Classification
Pan, Q., Zhang, Y., Chen, D., & Xu, G. (2017) · International Conference on Green Informatics (ICGI) · doi:10.1109/icgi.2017.31 · Cited by 18
“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 [38].”
BackgroundA review:deep learning technique for image classification
Aggarwal, V., & Kaur, G. (2018) · ACCENTS Transactions on Image Processing and Computer Vision · doi:10.19101/tipcv.2018.411003 · Cited by 17
“The process is done for all classes relying upon the sort of classification problem; binary classification or multi-class classification [7-10].”
BackgroundMethodTotal Recall: Understanding Traffic Signs Using Deep Convolutional Neural Network
Saha, S., Kamran, S. A., & Sabbir, A. S. (2018) · International Conference on Computer and Information Technology (ICCIT) · doi:10.1109/iccitechn.2018.8631925 · Cited by 22
“As the study [49], [50] suggests, accumulating more kernels cause performance degradation without provoking the high variance - high bias conundrum.”
BackgroundDeep Learning for Detecting Diseases in Gastrointestinal Biopsy Images
Srivastava, A. K., Sengupta, S., Kang, S., Kant, K., Khan, M. N., Ali, A., Moore, S. R., Amadi, B., Kelly, P. M., Syed, S., et al. (2019) · Systems and Information Engineering Design Symposium · doi:10.1109/sieds.2019.8735619 · Cited by 17
“Further it has been observed that accuracy of the models starts saturating by adding more layers after a certain point and then starts degrading at a fast rate [18].”
BackgroundDetecting Errors in Convolutional Neural Networks Using Inter Frame Spatio-Temporal Correlation
Draghetti, L. K., dos Santos, F. F., Carro, L., & Rech, P. (2019) · IEEE International On-Line Testing Symposium (IOLTS) · doi:10.1109/iolts.2019.8854431 · Cited by 17
“Even though researchers suggest there are ways to reduce CNN computational costs, it is still an issue for some application requirements [15].”
BackgroundA Brief Survey of Deep Learning Techniques for Person Re-identification
Mathur, N., Mathur, S., Mathur, D., & Dadheech, P. (2020) · International Conference on Emerging Technologies in Computer Engineering: Machine Learning and Internet of Things (ICETCE) · doi:10.1109/icetce48199.2020.9091747 · Cited by 17
“First, in this framework two convolutional neural networks (CNN) [45,59] are used which extracts both spatial and short-term motion features respectively from input videos.”
MethodSeq2Emoji: A hybrid sequence generation model for short text emoji prediction
Peng, D., & Zhao, H. (2021) · Knowledge-Based Systems · doi:10.1016/j.knosys.2020.106727 · Cited by 21
Multistage classification of oral histopathological images using improved residual network
Panigrahi, S., Bhuyan, R., Kumar, K., Nayak, J., & Swarnkar, T. (2021) · Mathematical Biosciences & Engineering · doi:10.3934/mbe.2022090 · Cited by 19
“It is observed from the literature that, convergence is inhibited and degradation happens as the number of layers in a CNN grows larger [48].”
BackgroundChen, Z. (2021) · Software Practice and Experience · doi:10.1002/spe.3066 · Cited by 17
Earthquake-Induced Building Damage Mapping Based on Multi-Task Deep Learning Framework
Chen, F., & Yu, B. (2019) · IEEE Access · doi:10.1109/access.2019.2958983 · Cited by 19
“Increasing the depth of the network often brings about higher training error [46].”
BackgroundOffline handwritten mathematical symbol recognition utilising deep learningPreprint
Nazemi, A., Tavakolian, N., Fitzpatrick, D., Fernando, C., & Suen, C. Y. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1910.07395 · Cited by 17
“This section initially presents the proposed method in this study for symbol segmentation followed by a description of all steps of utilising and modifying designed convolutional network for classification[23][24] over 100 classes in detail.”
Owais, M., Barhoum, E., & Younes, H. (2025) · Innovative Infrastructure Solutions · doi:10.1007/s41062-025-02421-z · Cited by 16
“The study number variable serves as a grouping identifier, allowing for an easy comparison of different experimental conditions.”
Multihop Neighbor Information Fusion Graph Convolutional Network for Text Classification
Lei, F., Liu, X., Li, Z., Dai, Q., & Wang, S. (2021) · Mathematical Problems in Engineering · doi:10.1155/2021/6665588 · Cited by 20
“Because the actual running time is sensitive to hardware and implementations, we follow He and Sun [45] to adopt the theoretical time complexity to show the complexity rather than the actual running time.”
MethodDeep Learning approximation of diffeomorphisms via linear-control systems
Scagliotti, A., & Scuola Internazionale Superiore di Studi Avanzati, T. (2022) · Mathematical Control and Related Fields · doi:10.3934/mcrf.2022036 · Cited by 16
“, [7]) and the accuracy of the network gets rapidly saturated (see [17]).”
BackgroundÖztürk, O., Ișık, M. S., Kada, M., & Şeker, D. Z. (2023) · Applied Sciences · doi:10.3390/app13106161 · Cited by 16
Data-Limited Modulation Classification With a CVAE-Enhanced Learning Model
Ji, X., Wang, J., Li, Y., Sun, Q., Jin, S., & Quek, T. Q. S. (2020) · IEEE Communications Letters · doi:10.1109/lcomm.2020.3004877 · Cited by 16
“In [12], the time complexity of all CONV layers in the classifier is described as (cid:5)(cid:6) (cid:7) , where n l is the number of filters in the l -th layer, s l is the filter size, and o l is the size of the output feature map.”
BackgroundResNet based on feature-inspired gating strategy
Miao, J., Shaowu, X., Zou, B., & Qiao, Y. (2021) · Multimedia Tools and Applications · doi:10.1007/s11042-021-10802-6 · Cited by 16
“In addition, considering some existing research results [6, 28], He et al. used 1 × 1 kernel for convolution as calculation method of the gating layer consistently, and Sigmoid instead of ReLU as the activation function.”
BackgroundResultEmoDNN: understanding emotions from short texts through a deep neural network ensemble
Kamran, S., Zall, R., Hosseini, S., Kangavari, M. R., Rahmani, S., & Hua, W. (2023) · Neural Computing and Applications · doi:10.1007/s00521-023-08435-x · Cited by 21
FCNN: An Efficient Intrusion Detection Method Based on Raw Network Traffic
Wang, Y., Jiang, Y., & Lan, J. (2021) · Security and Communication Networks · doi:10.1155/2021/5533269 · Cited by 17
Design Smells in Deep Learning Programs: An Empirical Study
Nikanjam, A., & Khomh, F. (2021) · IEEE International Conference on Software Maintenance (ICSM) · doi:10.1109/icsme52107.2021.00036 · Cited by 20
Hanussek, M., Bartusch, F., & Krüger, J. (2021) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1009244 · Cited by 17
“The time complexity of the training and testing procedures of a CNN is affected by six different parameters, the index of a convolutional layer, the number of such layers (depth), the number of filters (width) and their spatial size (length), the number of input channels, and the spatial size of the output feature map…”
Recurrent inference machines as inverse problem solvers for MR relaxometry
Sabidussi, E. R., Klein, S., Caan, M. W. A., Bazrafkan, S., den Dekker, A. J., Sijbers, J., Niessen, W. J., & Poot, D. H. J. (2021) · Medical Image Analysis · doi:10.1016/j.media.2021.102220 · Cited by 21
Al‐Bander, B., Yas, Q. M., Mahdi, H. F., & Al‐Hamd, R. K. S. (2021) · TURKISH JOURNAL OF ELECTRICAL ENGINEERING & COMPUTER SCIENCES · doi:10.3906/elk-2103-65 · Cited by 17
GSV-NET: A Multi-Modal Deep Learning Network for 3D Point Cloud Classification
Hoang, L., Lee, S., Lee, E.-J., & Kwon, K. (2022) · Applied Sciences · doi:10.3390/app12010483 · Cited by 16
Meta-Material Sensor Based Internet of Things: Design, Optimization, and Implementation
Hu, J., Zhang, H., Di, B., Han, Z., Poor, H. V., & Song, L. (2022) · IEEE Transactions on Communications · doi:10.1109/tcomm.2022.3187150 · Cited by 16
“Based on [43] and [44], the computational complexity of a convolutional/deconvolutional layer given a fixed kernel size can be calculated by where n in and n out denote the numbers of channels of the input and output data, and m 1 and m 2 denote the side-lengths of output data.”
BackgroundImpatient DNNs - Deep Neural Networks with Dynamic Time Budgets
Amthor, M., Rodner, E., & Denzler, J. (2016) · British Machine Vision Conference (BMVC) · doi:10.5244/c.30.116 · Cited by 22
Uncertainty-Aware Gradient Stabilization for Small Object Detection
Sun, H., Li, Y., Yang, L., Cao, X., & Zhang, B. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.00787 · Cited by 28
“Recent advances in deep neural networks (DNNs) [16, 59] have substantially improved the object detection field [11, 33].”
ThumbNet: One Thumbnail Image Contains All You Need for Recognition
Zhao, C., & Ghanem, B. (2020) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3394171.3413937 · Cited by 15
“into the question of how to speed up deep networks, let us first analyze what dominates the computational complexity of a CNN. We calculate the total time complexity of convolutional layers as follows [12]: O X d l=1 n l l1 s 2 n l m 2 ; (1) where lis the index of a layer and dis the total number of layers; n l 1 is th…”
BackgroundEffective Artificial Neural Network Framework for Time-Modulated Arrays Synthesis
Hei, Y. Q., Yuan, L., Li, W. T., Mou, J. C., & Shi, X. (2023) · IEEE Transactions on Antennas and Propagation · doi:10.1109/tap.2023.3303464 · Cited by 15
Design of a Chaotic Communication System Based on Deep Learning With Two-Dimensional Reshaping
He, L., Lai, R., Shao, S., & Li, Z. (2024) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2024.3382625 · Cited by 16
Cai, M., Chen, S., Liu, J., Yang, Y., & Cen, L. (2024) · Journal of Intelligent Manufacturing · doi:10.1007/s10845-024-02401-7 · Cited by 17
“The TC and SC of the GAT layer are given by He and Sun (2015) and are defined as where F is the dimension of the input features, F is the dimension of the features obtained by linear transformation, and | V | and | E | are the number of nodes and edges in the graph, respectively.”
Xu, Y., Lu, W., Pan, Z., Wang, Z., Luo, C., & Bai, Y. (2024) · Journal of Hydrology · doi:10.1016/j.jhydrol.2024.131904 · Cited by 15
Ding, A., Zhang, Y., Zhu, L., Du, Y., & Ma, L. (2020) · Measurement · doi:10.1016/j.measurement.2020.107942 · Cited by 18
Bridging Adversarial and Statistical Domain Transfer via Spectral Adaptation Networks
Raab, C., Väth, P., Meier, P., & Schleif, F.-M. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-69535-4_28 · Cited by 16
“Given the computational complexity of convolution [29] and df = 256 ≪ dx in ASAN, the computation of SVD at the bottleneck layer does not increase the complexity class of the network.”
MethodRecognition of partially occluded faces using regularized ICA
Maghari, A. Y. A. (2020) · Inverse Problems in Science and Engineering · doi:10.1080/17415977.2020.1845329 · Cited by 17
Liu, X., Xia, G., Lei, F., Zhang, Y., & Chang, S. (2021) · IEEE Access · doi:10.1109/access.2021.3060173 · Cited by 17
“Because the actual running time is sensitive to hardware and implementations, we adopt the theoretical time complexity (instead of the actual running time) to show the complexity, VOLUME 9, 2021 following He and Sun [53].”
MethodIoT System Based on parameter optimization of Deep Learning using Genetic Algorithm
Slim, S. O., Elfattah, M. M. A., Atia, A., & Mostafa, M.-S. (2021) · International journal of intelligent engineering and systems · doi:10.22266/ijies2021.0430.20 · Cited by 16
“The training is approximately three times testing time (forward propagation takes one and backward propagation takes two) [46].”
BackgroundMethodDriver Activity Recognition Through Deep Learning
Nel, F., & Ngxande, M. (2021) · Southern African Universities Power Engineering Conference/Robotics and Mechatronics/Pattern Recognition Association of South Africa (SAUPEC/RobMech/PRASA) · doi:10.1109/saupec/robmech/prasa52254.2021.9377022 · Cited by 15
“In [24], the authors explain that such degradation is not a result of overfitting, as the training errors also worsen.”
BackgroundKadulkar, S., Howard, M. P., Truskett, T. M., & Ganesan, V. (2021) · The Journal of Physical Chemistry B · doi:10.1021/acs.jpcb.1c02004 · Cited by 17
Adversarial Learning-Based Spectrum Sensing in Cognitive Radio
Wang, C., Xu, Y., Chen, Z., Tian, J., Cheng, P., & Li, M. (2021) · IEEE Wireless Communications Letters · doi:10.1109/lwc.2021.3133883 · Cited by 15
Tu, Y.-K., Kuo, C., Fang, S.-L., Chen, H., Chi, M.-K., Yao, M.-H., & Kuo, B. (2022) · Agriculture · doi:10.3390/agriculture12020259 · Cited by 18
“However, gradient exploding or vanishing and degradation problems may arise in the deep architecture of the neural network in CNN-based models, and this may impede the model training process and reduce the performance of the CNN model [54,55].”
BackgroundMulti‐step implicit Adams predictor‐corrector network for fire detection
Deng, Z., Hu, S., Yin, S., Wang, Y., Basu, A., & Cheng, I. (2022) · IET Image Processing · doi:10.1049/ipr2.12491 · Cited by 16
Iyer, S., Blair, A., White, C., Dawes, L. C., Moses, D. A., & Sowmya, A. (2023) · Informatics in Medicine Unlocked · doi:10.1016/j.imu.2023.101238 · Cited by 19
Deep Learning Ultrasound Computed Tomography Under Sparse Sampling
Long, X., Chen, J., Liu, W., & Tian, C. (2023) · IEEE Transactions on Ultrasonics Ferroelectrics and Frequency Control · doi:10.1109/tuffc.2023.3299954 · Cited by 15
“tional operation lies in the convolutional layers [52], given by”
BackgroundOPEX-Limited 5G RAN Slicing: an Over-Dataset Constrained Deep Learning Approach
Chergui, H., & Verikoukis, C. V. (2020) · International Conference on Communications · doi:10.1109/icc40277.2020.9149051 · Cited by 14
“In this regard, we notice that this CNN strategy is of high complexity [9].”
BackgroundConstrained Center Loss for Convolutional Neural Networks
Shi, Z.-L., Wang, H., & Leung, C.-S. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3104392 · Cited by 15
“The computational complexity for backward propagation is about two times of forward propagation [50].”
BackgroundDeep Learning-based Signal Detection for Uplink in LoRa-like Networks
Tesfay, A. A., Simon, E. P., Kharbech, S., & Clavier, L. (2021) · IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) · doi:10.1109/pimrc50174.2021.9569470 · Cited by 20
Augmenting Deep Classifiers with Polynomial Neural Networks
Chrysos, G. G., Georgopoulos, M., Deng, J., Kossaifi, J., Panagakis, Y., & Anandkumar, A. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19806-9_40 · Cited by 34
A Study on Path Loss Modeling using ResNet and Pre-Training with Free Space Path Loss
Nagao, T., & Hayashi, T. (2022) · IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) · doi:10.1109/pimrc54779.2022.9977760 · Cited by 15
“However, as layers become deeper, such structures have limitations in improving training error, known as the degradation and vanishing gradient problems [15].”
BackgroundRecent advances in agricultural disease image recognition technologies: A review
Bondre, S., & Patil, D. (2023) · Concurrency and Computation Practice and Experience · doi:10.1002/cpe.7644 · Cited by 19
Fast and Reliable Score-Based Generative Model for Parallel MRI
Hou, R., Li, F., & Zeng, T. (2023) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2023.3333538 · Cited by 17
Deep learning algorithms for hyperspectral remote sensing classifications: an applied review
Pal, M. (2024) · International Journal of Remote Sensing · doi:10.1080/01431161.2023.2297178 · Cited by 15
“On the other hand, studies by He and Sun (2015) and Srivastava, Greff, and Schmidhuber (2015) suggested that increasing depth of the network forces classification accuracy to saturate followed by degradation.”
BackgroundBenoit, A., Escriba, C., Gauchard, D., Estève, A., & Rossi, C. (2024) · IEEE Sensors Journal · doi:10.1109/jsen.2024.3364249 · Cited by 20
“According to He and Sun [74], the time complexity of convolution layers is estimated by O ( di = 1 f i − 1 s 2 i f i m 2 i ) , where d is the number of convolutional layers, f i is the number of filters in the i th layer, f i − 1 is the number of input channels of the ith layer, s i is the spatial…”
BackgroundHoang, D. M. T., Pham, T. V., Pham, A. T., & Nguyen, C. T. (2024) · IEEE Access · doi:10.1109/access.2024.3412055 · Cited by 14
“According to [46], [47], and [48], An intuitive explanation for this behavior is that when δ decreases and/or ζ increases, to keep maximizing the reward defined in (23), the BER of Bob’s channel p e , B is forced to decrease as well.”
BackgroundFast Object Detection at Constrained Energy
Liu, J., Huang, Y., Peng, J., Yao, J., & Wang, L. (2016) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2016.2577538 · Cited by 14
Shallow Convolutional Neural Networks for Acoustic Scene Classification
Lu, L., Yang, Y., Jiang, Y., Ai, H., & Tu, W. (2018) · Wuhan University Journal of Natural Sciences · doi:10.1007/s11859-018-1308-z · Cited by 14
A Hybrid-3D Convolutional Network for Video Compressive Sensing
Zhao, Z., Xie, X., Liu, W., & Pan, Q. (2020) · IEEE Access · doi:10.1109/access.2020.2969290 · Cited by 16
“We also calculate the theoretical time complexity [59] of the data-driven networks.”
BackgroundVaddi, R., & Manoharan, P. (2020) · IET Image Processing · doi:10.1049/iet-ipr.2020.0728 · Cited by 14
Lu, H. (2021) · BioMed Research International · doi:10.1155/2021/5513746 · Cited by 14
“He and Sun [10] discussed how to balance the number of layers, the number of feature maps, and the size of the convolution kernel in the limited training time and computational complexity; they showed that the recognition performance of CNN with small convolution kernel and deep layers is better than that with large c…”
BackgroundSemantic Service Clustering With Lightweight BERT-Based Service Embedding Using Invocation Sequences
Zeng, K., & Paik, I. (2021) · IEEE Access · doi:10.1109/access.2021.3069509 · Cited by 15
“The time complexity generally represents multiply-accumulate operations [42], [43].”
BackgroundConvolutional neural network architectures
Huang, S.-C., & Le, T.-H. (2021) · Elsevier eBooks · doi:10.1016/b978-0-323-90198-7.00001-x · Cited by 15
Reconstructing cosmic polarization rotation with ResUNet-CMB
Guzman, E., & Meyers, J. (2022) · Journal of Cosmology and Astroparticle Physics · doi:10.1088/1475-7516/2022/01/030 · Cited by 16
Shi, Z., Liu, X., Chen, J., Zi, Y., & Zhou, Z. (2022) · ISA Transactions · doi:10.1016/j.isatra.2022.01.011 · Cited by 14
Pathak, D., & Raju, U. S. N. (2022) · Concurrency and Computation Practice and Experience · doi:10.1002/cpe.6851 · Cited by 19
GAN-rcLSTM: A Deep Learning Model for Radar Echo Extrapolation
Geng, H., Wang, T., Zhuang, X., Du, X., Hu, Z., & Geng, L. (2022) · Atmosphere · doi:10.3390/atmos13050684 · Cited by 15
A Survey for Sparse Regularization Based Compression Methods
Tang, A., Quan, P., Niu, L., & Shi, Y. (2022) · Annals of Data Science · doi:10.1007/s40745-022-00389-6 · Cited by 20
“For example, if a convolution network has l convolution layers, the number of filters in layer l is nl , the size of the filter is sl , the size of the active graph isml , the sum of the computational complexity of all the convolution layers can be denoted as O( ∑L l=1 nl−1 · s2 l · nl ·m2 l ) [19].”
BackgroundTesfay, A. A., Kharbech, S., Simon, E. P., & Clavier, L. (2022) · IEEE Communications Letters · doi:10.1109/lcomm.2022.3217337 · Cited by 14
“As a series of convolutional layers, and based on [16], the complexity order of the proposed detection scheme is estimated to be O ( N 4 ) + O ( N 2 M ) .”
BackgroundVerbickas, R., Laganière, R., Laroche, D., Zhu, C., Xu, X., & Ors, A. O. (2017) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw.2017.62 · Cited by 15
“Strategy 3 is based on observations made in prior experimental results [17], which attempts to keep feature maps relatively large (to the input) in an attempt to learn better features.”
BackgroundEfficient convNets for fast traffic sign recognition
Luo, X., Zhu, J., & Yu, Q. (2019) · IET Intelligent Transport Systems · doi:10.1049/iet-its.2018.5489 · Cited by 14
“In [16], experiments show that overly increasing depth can harm the accuracy, even if the width is unchanged.”
BackgroundObject Detection Using Vision and LiDAR Sensor Fusion for Multi-channel V2X System
Lee, G. H., Choi, J. D., Lee, J. H., & Kim, M. Y. (2020) · International Conference on Artificial Intelligence in Information and Communication (ICAIIC) · doi:10.1109/icaiic48513.2020.9065243 · Cited by 15
“In this paper, we use a CNN based detection algorithm specialized for image processing for object classification[6].”
MethodTowards End-to-End Deep Learning Performance Analysis of Electric Motors
Gabdullin, N., Madanzadeh, S., & Vilkin, A. (2021) · Actuators · doi:10.3390/act10020028 · Cited by 14
“ResNets [40] are deep residual CNNs that were developed to explore the relation between the depth and image classification accuracy inspired by the evidence of accuracy degradation for extremely deep architectures [41].”
BackgroundLightweight Network Design Based on ResNet Structure for Modulation Recognition
Lu, X. Q., Tao, M., Fu, X. G., Gui, G., Ohtsuki, T. O., & Sari, H. (2021) · IEEE Vehicular Technology Conference · doi:10.1109/vtc2021-fall52928.2021.9625558 · Cited by 16
Shallow and wide fractional max-pooling network for image classification
Yue, K., Xu, F., & Yu, J. (2017) · Neural Computing and Applications · doi:10.1007/s00521-017-3073-x · Cited by 15
“The problem of vanishing/exploding gradients and degradation is found in [3, 9], which not only hampers convergence from the beginning, but also results in hard training of deep net and leads to a higher training error as reported in [15, 30].”
Proximal Parameter Distribution Optimization
Wang, X., Li, T., & Cheng, Y. (2019) · IEEE Transactions on Systems Man and Cybernetics Systems · doi:10.1109/tsmc.2019.2931946 · Cited by 14
“Mnih et al. [8] proposed a novel deep RL method, i.e., deep Q network (DQN) [9]–[12] by combining deep convolutional network [13]–[15] with traditional Q learning [1], which successfully enabled agents to achieve or exceed human-level in Atari games [16].”
MethodExploring the Design Space of Deep Convolutional Neural Networks at Large ScalePreprint
Iandola, F. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1612.06519 · Cited by 23
“He and H. Sun applied delayed downsampling to four different CNN architectures, and in each case delayed downsampling led to higher classification accuracy [78].”
BackgroundMethodInfluential
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.