The DOI matches its Crossref record; 144 of the 157 references it cites were found in a registry or an index.
Efficient Processing of Deep Neural Networks: A Tutorial and Survey
Abstract
Deep neural networks (DNNs) are currently widely used for many artificial intelligence (AI) applications including computer vision, speech recognition, and robotics. While DNNs deliver state-of-the-art accuracy on many AI tasks, it comes at the cost of high computational complexity. Accordingly, techniques that enable efficient processing of DNNs to improve energy efficiency and throughput without sacrificing application accuracy or increasing hardware cost are critical to the wide deployment of DNNs in AI systems. This article aims to provide a comprehensive tutorial and survey about the recent advances toward the goal of enabling efficient processing of DNNs. Specifically, it will provide an overview of DNNs, discuss various hardware platforms and architectures that support DNNs, and highlight key trends in reducing the computation cost of DNNs either solely via hardware design changes or via joint hardware design and DNN algorithm changes. It will also summarize various development resources that enable researchers and practitioners to quickly get started in this field, and highlight important benchmarking metrics and design considerations that should be used for evaluating the rapidly growing number of DNN hardware designs, optionally including algorithmic codesigns, being proposed in academia and industry. The reader will take away the following concepts from this article: understand the key design considerations for DNNs; be able to evaluate different DNN hardware implementations with benchmarks and comparison metrics; understand the tradeoffs between various hardware architectures and platforms; be able to evaluate the utility of various DNN design techniques for efficient processing; and understand recent implementation trends and opportunities.
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
“Classical arithmetic operations, such as MAC, constitute the bulk of the computational requirements for models used in artificial intelligence [13].”
From Brain-inspired computing systems: a systematic literature review (2024) · cites this paper as [13]
In the citing paper
“To clarify, consider the multiply-and-accumulate (MAC) operation in digital processors, where the operation energy consumption is significantly dependent on the bit width of the input operands [13, 235] or equivalently to the signal-to-noise ratio (SNR) in analog computing [236].”
From Brain-inspired computing systems: a systematic literature review (2024) · cites this paper as [13]
In the citing paper
“Altogether, processing one 224 × 224 input image needs a VGG-16 model of 15.5 G multiply-and-accumulates and 138 M weights [40].”
From Face Recognition Based on Deep Learning and FPGA for Ethnicity Identification (2022) · cites this paper as [40]
In the citing paper
“The energy consumption of executing a DNN model is highly dictated by the computation architecture and methods (e.g., ultra-low power compute-in-memory AI accelerator) at the edge computation nodes [249], the architecture of DNN models [250], and the wireless transmission for data exchange during the model inference procedure.”
From Edge artificial intelligence for 6g: vision, enabling technologies, and applications (2022) · cites this paper as [250]
In the citing paper
“but there is no review discussing all three categories which in future will be integrated seamlessly to design effective and efficient DNNs in the new edge intelligence era.”
From Bringing ai to edge: from deep learning’s perspective (2022) · cites this paper as [25]
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 15×
“In 2015, the ImageNet winning entry, ResNet [15], exceeded human-level accuracy with a top-5 error rate4 below 5%.”
InfluentialVery 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 10×
“Used in VGG-16. decompose 5x5 filter 5x1 filter 1x5 filter Apply sequentially (b) Constructing a 5×5 support from 1×5 and 5×1 filter.”
InfluentialImageNet 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 10×
“this, local response normalization (LRN) [3] was used, which”
BackgroundMethodInfluentialHochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318 · cited 1×
“In contrast, recurrent neural networks (RNNs), of which Long Short-Term Memory networks (LSTMs) [38] are a popular variant, have internal memory to allow long-term dependencies to affect the output.”
LeCun, Y., Bengio, Y., & Hinton, G. (2015) · Nature · doi:10.1038/nature14539 · Cited by 84,126
Gradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805 · cited 4×
“Table 2 Summary of Popular DNNs [3], [15], [48], [50], [51]”
InfluentialGoing 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 · cited 8×
“†ACCURACY IS MEASURED BASED ON TOP-5 ERROR ON IMAGENET [14].”
InfluentialFully convolutional networks for semantic segmentation
Long, J., Shelhamer, E., & Darrell, T. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298965 · Cited by 38,834
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
Research works printed but not linked: 34
- 10.13140/RG.2.2.18893.74727 · DOI · Paper page
- Li, Stanford CS Class CS231n: Convolutional Neural Networks for Visual Recognition
- 2016, Complete Visual Networking Index (VNI) Forecast
- Woodhouse, 2016, Big, Big, Big Data: Higher and Higher Resolution Video Surveillance
- Maas, Rectifier nonlinearities improve neural network acoustic models, 1
- Caffe LeNet MNIST
- Caffe Model Zoo
- Matconvnet Pretrained Models
- TensorFlow-Slim Image Classification Library
- Deep Learning Frameworks
- Yann, The MNIST DATABASE of Handwritten Digits
- Krizhevsky, The CIFAR-10 Dataset
- Krizhevsky, 2010, Convolutional deep belief networks on cifar-10, Unpublished Manuscript, 40, 1
- Pascal VOC Data Sets
- Microsoft Common Objects in Context (COCO) Dataset
- Google Open Images
- YouTube-8M
- AudioSet
- 2017, Intel Architecture Instruction Set Extensions Programming Reference
- Condon, Facebook unveils Big Basin, new server geared for deep learning
- Intel Math Kernel Library
- 10.1145/1816038.1815993 · DOI
- 10.1145/2872887.2750389 · DOI
- 10.1145/2644865.2541967 · DOI · Paper page
- Standard, High bandwidth memory (HBM) DRAM
- 10.1145/3093336.3037702 · DOI
- Wilson, International technology roadmap for semiconductors (ITRS)
- 10.1109/icvd.2004.1260946 · doi:10.1109/icvd.2004.1260946
- Lee, A 2.5 GHz 7.7 TOPS/W switched-capacitor matrix multiplier with co-designed local memory in 40 nm, 418
- Lee, LogNet: Energy-efficient neural networks using logrithmic computations, 5900
- DNN Energy Estimation
- 10.1145/3140659.3080254 · DOI · Paper page
- Iandola, SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1 MB model size
- Benchmarking DNN Processors
Cited by
Items 601–700 of 1,011
Autonomous Driving with Deep Learning: A Survey of State-of-Art TechnologiesPreprint
Huang, Y., & Chen, Y. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.06091 · Cited by 115
“So far there are some techniques proposed for use [21, 22], roughly categorized into four categories: parameter pruning and sharing, low-rank factorization, transferred/compact convolutional filters, and knowledge distillation .”
MethodSelf-citationBrain-inspired computing systems: a systematic literature review
Zolfagharinejad, M., Alegre-Ibarra, U., Chen, T., Kinge, S., & van der Wiel, W. G. (2024) · The European Physical Journal B · doi:10.1140/epjb/s10051-024-00703-6 · Cited by 39
Breath VOC analysis and machine learning approaches for disease screening: a review
P, H., Rangarajan, M., & Pandya, H. J. (2023) · Journal of Breath Research · doi:10.1088/1752-7163/acb283 · Cited by 37
Analog Photonics Computing for Information Processing, Inference, and Optimization
Stroev, N., & Berloff, N. G. (2023) · Advanced Quantum Technologies · doi:10.1002/qute.202300055 · Cited by 59
Paithane, P. M., & Kakarwal, S. (2023) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.121064 · Cited by 38
Hypermultiplexed integrated photonics–based optical tensor processor
Ou, S., Xue, K., Zhou, L., Lee, C., Sludds, A., Hamerly, R., Zhang, K., Feng, H., Yu, Y., Kopparapu, R., et al. (2025) · Science Advances · doi:10.1126/sciadv.adu0228 · Cited by 52
“Because of the massive memory interface requirements in tensor processing ( 12 ), traditional Von Neumann architectures are inefficient in performing these tasks.”
Towards Trustworthy AI: A Review of Ethical and Robust Large Language Models
Ferdaus, M. M., Abdelguerfi, M., Ioup, E., Niles, K. N., Pathak, K., & Sloan, S. D. (2025) · ACM Computing Surveys · doi:10.1145/3777382 · Cited by 136
Hardware-Software Co-Design for an Analog-Digital Accelerator for Machine Learning
Ambrosi, J., Ankit, A., Antunes, R., Chalamalasetti, S. R., Chatterjee, S., Hajj, I. E., Fachini, G. J. D. A., Faraboschi, P., Foltín, M., Huang, S., et al. (2018) · IEEE International Conference on Rebooting Computing (ICRC) · doi:10.1109/icrc.2018.8638612 · Cited by 47
“A comprehensive survey and categorization of these works has been done by Sze et al. [68].”
BackgroundKwon, O., Kim, H. G., Kim, W., Kim, G.-H., & Kim, K. (2020) · IEEE Access · doi:10.1109/access.2020.2970026 · Cited by 37
“high-level features such as shapes combined by lines and finally, DNN provides the probability to infer a certain object from these feature information [11].”
BackgroundMethodMedeiros, D. S. V., Neto, H. N. C., Lopez, M. A., Magalhães, L., Fernandes, N. C., Vieira, A. B., Silva, E. F., & Mattos, D. M. F. (2020) · Journal of Internet Services and Applications · doi:10.1186/s13174-020-00127-2 · Cited by 46
“In the unsupervised learning process, there is no labeled data and, therefore, the main goal is to generate labeled data using unsupervised learning algorithms, such as Restricted Boltzmann Machine (RBM), Deep Belief Network (DBN), Deep Neural Network (DNN), generalized AutoEncoders, and Recurrent Neural Network (RNN)…”
BackgroundKim, J. K., Choo, Y. J., & Chang, M. C. (2021) · Journal of Stroke and Cerebrovascular Diseases · doi:10.1016/j.jstrokecerebrovasdis.2021.105856 · Cited by 44
Optimally Scheduling CNN Convolutions for Efficient Memory AccessPreprint
Stoutchinin, A., Conti, F., & Benini, L. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1902.01492 · Cited by 57
Technology Prospects for Data-Intensive Computing
Akarvardar, K., & Wong, H. ‐. P. (2023) · Proceedings of the IEEE · doi:10.1109/jproc.2022.3218057 · Cited by 42
“For instance, neural networks are inherently suited for hardware parallelization, since their computation is primarily based on multiplyand-accumulate (MAC) operations [10].”
BackgroundZhang, P., Huang, W., Chen, Y., Zhou, M., & Al‐Turki, Y. A. (2023) · IEEE Transactions on Automation Science and Engineering · doi:10.1109/tase.2023.3244184 · Cited by 37
“With the rapid development of deep learning and the continuous upgrading of computing environments, deep neural network (DNN) technologies are constantly developing and have significantly impacted many fields, such as computer vision, data mining, and natural language processing [9].”
BackgroundSelf-citationYuan, S., Ajam, H., sinnah, Z. A. B., Altalbawy, F. M. A., AbdulAmeer, S. A., Husain, A., Mashhadani, Z. I. A., Alkhayyat, A. H., Alsalamy, A. H., Zubaid, R. A., et al. (2023) · Ecotoxicology and Environmental Safety · doi:10.1016/j.ecoenv.2023.115066 · Cited by 38
TransAct: Transformer-based Realtime User Action Model for Recommendation at Pinterest
Xia, X.-G., Eksombatchai, P., Pancha, N., Badani, D., Wang, P.-W., Gu, N., Joshi, S. V., Farahpour, N., Zhang, Z., & Zhai, A. (2023) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3580305.3599918 · Cited by 66
“However, for GPU-based inference, larger batches are more efficient [29].”
Trading-Off Accuracy and Energy of Deep Inference on Embedded Systems: A Co-Design Approach
Jayakodi, N. K., Chatterjee, A., Choi, W., Doppa, J. R., & Pande, P. P. (2018) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2018.2857338 · Cited by 39
“also have an overhead of analog to digital conversion [6].”
BackgroundMethodSurvey of Precision-Scalable Multiply-Accumulate Units for Neural-Network Processing
Camus, V., Enz, C., & Verhelst, M. (2019) · IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS) · doi:10.1109/aicas.2019.8771610 · Cited by 36
“Many hardware improvements have recently been proposed, innovating at different abstraction levels to solve both memory and computational bottlenecks [1], [2].”
BackgroundA CNN-based approach to classify cricket bowlers based on their bowling actions
Islam, M. N. A., Hassan, T. B., & Khan, S. K. (2019) · IEEE International Conference on Signal Processing, Information, Communication and Systems (SPICSCON) · doi:10.1109/spicscon48833.2019.9065090 · Cited by 36
“Convolutional neural network (CNN) [11, 12] is a deep learning architecture that has been massively fruitful for image classification [13].”
BackgroundAdvantages and Limitations of Fully on-Chip CNN FPGA-Based Hardware Accelerator
Dinelli, G., Meoni, G., Rapuano, E., & Fanucci, L. (2020) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas45731.2020.9180867 · Cited by 38
A Survey of Deep Learning for Data Caching in Edge Network
Wang, Y., & Friderikos, V. (2020) · Informatics · doi:10.3390/informatics7040043 · Cited by 45
“Recently, deep learning (DL) has attracted significant attention from both academia and industry and has been applied to diverse domains like self-driving, medical diagnosis, playing complex games such as Go [2].”
BackgroundPointAR: Efficient Lighting Estimation for Mobile Augmented Reality
Zhao, Y., & Guo, T. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58592-1_40 · Cited by 36
Real-Time SSDLite Object Detection on FPGA
Kim, S., Na, S., Kong, B. Y., Choi, J., & Park, I. (2021) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2021.3064639 · Cited by 42
Shim, S., Kim, J., Cho, G.-C., & Lee, S.-W. (2022) · Structural Health Monitoring · doi:10.1177/14759217221097868 · Cited by 40
Egger, J., Pepe, A., Gsaxner, C., Jin, Y., Li, J., & Kern, R. (2021) · PeerJ Computer Science · doi:10.7717/peerj-cs.773 · Cited by 52
INTRUSION DETECTION IN COMPUTER NETWORKS USING LATENT SPACE REPRESENTATION AND MACHINE LEARNING
Hamolia, V., Melnyk, V. A., Zhezhnych, P., & Shilinh, A. (2020) · International Journal of Computing · doi:10.47839/ijc.19.3.1893 · Cited by 38
Low-power deep learning edge computing platform for resource constrained lightweight compact UAVs
Albanese, A., Nardello, M., & Brunelli, D. (2022) · Sustainable Computing Informatics and Systems · doi:10.1016/j.suscom.2022.100725 · Cited by 60
Pineapples’ Detection and Segmentation Based on Faster and Mask R-CNN in UAV Imagery
Shiu, Y.-S., Lee, R.-Y., & Chang, Y.-C. (2023) · Remote Sensing · doi:10.3390/rs15030814 · Cited by 40
A 2941-TOPS/W Charge-Domain 10T SRAM Compute-in-Memory for Ternary Neural Network
Cheon, S., Lee, K., & Park, J. (2023) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2023.3241385 · Cited by 40
“The inputs and weights from different channels are mapped to be processed in the same array, utilizing the weight-stationary [31] for data reuse.”
MethodIyer, V., Issadore, D., & Aflatouni, F. (2023) · Lab on a Chip · doi:10.1039/d2lc01163h · Cited by 38
Prabakaran, B. S., Hamelmann, P., Ostrowski, E. J., & Shafique, M. (2023) · IEEE Access · doi:10.1109/access.2023.3284315 · Cited by 48
“, 2021), to further reduce the model’s size and improve its computational performance; INT8 computations are several orders of magnitude faster than 32-bit floating-point (FP32) operations (Sze et al., 2017).”
BackgroundJiang, Y., Wang, W., Zou, L., & Cao, Y. (2023) · Acta Geotechnica · doi:10.1007/s11440-023-01950-0 · Cited by 37
A Low Power and Low Latency FPGA-Based Spiking Neural Network Accelerator
Liu, H., Chen, Y., Zeng, Z., Zhang, M., & Qu, H. (2023) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn54540.2023.10191153 · Cited by 38
“Changing from floating point to fixed point with bit-width reduce can decrease the energy and area cost for storage, since memory access and data movement dominate energy consumption [34].”
BackgroundSelf-citationCombining machine learning algorithms for personality trait prediction
Serrano‐Guerrero, J., Alshouha, B., Bani‐Doumi, M., Chiclana, F., Romero, F. P., & Olivas, J. Á. (2024) · Egyptian Informatics Journal · doi:10.1016/j.eij.2024.100439 · Cited by 44
Jafary, P., Shojaei, D., Rajabifard, A., & Ngo, T. (2024) · Cities · doi:10.1016/j.cities.2024.105115 · Cited by 35
Li, Y., Fu, B., Yin, Y., Hu, X., Wang, W., Wang, W., Li, X. S., & Long, G. (2024) · Intelligent geoengineering. · doi:10.1016/j.ige.2024.10.003 · Cited by 40
Machine Learning Applications in Energy Harvesting Internet of Things Networks: A Review
Alamu, O., Olwal, T. O., & Migabo, E. (2025) · IEEE Access · doi:10.1109/access.2024.3525263 · Cited by 48
Zhang, H., Lee, H.-J., & Ko, S. (2018) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas.2018.8351354 · Cited by 37
“Many research works [2] have been done recently in order to reduce the hardware cost.”
BackgroundTime–frequency masking based supervised speech enhancement framework using fuzzy deep belief network
Samui, S., Chakrabarti, I., & Ghosh, S. K. (2018) · Applied Soft Computing · doi:10.1016/j.asoc.2018.10.031 · Cited by 36
Complexity-Scalable Neural-Network-Based MIMO Detection With Learnable Weight Scaling
Mohammad, A., Masouros, C., & Andreopoulos, Y. (2020) · IEEE Transactions on Communications · doi:10.1109/tcomm.2020.3007622 · Cited by 42
Baek, J., & Choi, Y. (2019) · Applied Sciences · doi:10.3390/app9194180 · Cited by 38
“Deep learning is an artificial intelligence technology that learns object characteristics hierarchically, without human interference, and improves the prediction performance by means of error evaluation [26].”
BackgroundPreVIous: A Methodology for Prediction of Visual Inference Performance on IoT Devices
Velasco-Montero, D., Fernández‐Berni, J., Carmona-Galan, R., & Rodriguez-Vazquez, A. (2020) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.2981684 · Cited by 43
“The success of DL in enabling practical vision algorithms and unifying the procedure for a number of tasks such as image recognition, object detection, and pixel segmentation, has prompted research and development at various levels [5].”
BackgroundDSP-Efficient Hardware Acceleration of Convolutional Neural Network Inference on FPGAs
Wang, D., Xu, K., Guo, J., & Ghiasi, S. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.2968023 · Cited by 38
“CNNs are generally composed of multiple functional layers [2], each of which performs a certain type of arithmetic operation on the input image or feature”
BackgroundWhen RSSI encounters deep learning: An area localization scheme for pervasive sensing systems
Shen, Z., Zhang, T., Tagami, A., & Jin, J. (2020) · Journal of Network and Computer Applications · doi:10.1016/j.jnca.2020.102852 · Cited by 35
Test and Yield Loss Reduction of AI and Deep Learning Accelerators
Sadi, M., & Guin, U. (2021) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2021.3051841 · Cited by 38
Handwriting-based gender and handedness classification using convolutional neural networks
Rahmanian, M., & Shayegan, M. A. (2021) · Multimedia Tools and Applications · doi:10.1007/s11042-020-10170-7 · Cited by 40
Photonic Integrated Reconfigurable Linear Processors as Neural Network Accelerators
Marinis, L. D., Cococcioni, M., Liboiron-Ladouceur, O., Contestabile, G., Castoldi, P., & Andriolli, N. (2021) · Applied Sciences · doi:10.3390/app11136232 · Cited by 39
“Computer vision is probably the most relevant and wide application scenario in which deep learning has become a very successful and currently standard approach [1].”
SPACX: Silicon Photonics-based Scalable Chiplet Accelerator for DNN Inference
Li, Y., Louri, A., & Karanth, A. (2022) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca53966.2022.00066 · Cited by 36
“However, the highly regular and non-uniform communication in DNN inference applications [1], [15], [16] makes the costly equal bandwidth allocation unnecessary.”
BackgroundMethodInfluentialEncoding Spectral and Spatial Context Information for Hyperspectral Image Classification
Sun, X., Zhou, F., Dong, J., Gao, F., Mu, Q., & Wang, X. (2017) · IEEE Geoscience and Remote Sensing Letters · doi:10.1109/lgrs.2017.2759168 · Cited by 35
Cao, T., Liu, C., Wang, W., Zhang, T., Lee, H. K., Li, M. H., Song, W., Chen, Z. X., Zhuo, V. Y.-Q., Wang, N., et al. (2022) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2022.3214334 · Cited by 34
“To speed up the computation, the most time consuming VMM operations are implemented in NVIDA CUDA and processed in GPU.”
MethodReconfigurability, Why It Matters in AI Tasks Processing: A Survey of Reconfigurable AI Chips
Wei, S., Lin, X., Tu, F., Wang, Y., Liu, L., & Yin, S. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3228860 · Cited by 40
“Some of them are written from a general perspective, such as the papers proposed by Joel Emer’s group [3], Yiran Chen and Yuan Xie’s groups [6], and a multinational group [7].”
BackgroundSlope stability analysis of heavy-haul freight corridor using novel machine learning approach
Sabri, M. S., Ahmad, F., & Samui, P. (2023) · Modeling Earth Systems and Environment · doi:10.1007/s40808-023-01774-7 · Cited by 34
“Every node in the other layers of a DNN creates a fully integrated neural sys-tem, and each node in a layer is assigned a certain weight, denoted as w ij (Sze et al. 2017).”
InfluentialContent-Adaptive Downsampling in Convolutional Neural Networks
Hesse, R., Schaub-Meyer, S., & Roth, S. (2023) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw59228.2023.00478 · Cited by 52
“As our approach improves the cost-accuracy trade-off of various established models, it contributes to saving valuable scarce resources and to the important research direction of more (energy) efficient deep learning [40].”
BackgroundEffective DDoS Mitigation via ML-Driven In-Network Traffic Shaping
Zhao, Z., Liu, Z., Chen, H., Zhang, F., Song, Z., & Li, Z. (2024) · IEEE Transactions on Dependable and Secure Computing · doi:10.1109/tdsc.2023.3349180 · Cited by 37
“However, directly executing ML inferences as auser-spaceapplicationis knowntobeslow onCPUs due to a large number of multiply and accumulate (MAC) operations ( e.g., VGG and ResNet50 need 1.4 and 3.9 billion MACs, respectively [35]).”
BackgroundInfluentialManzak, D., Çetinel, G., & Manzak, A. (2019) · International Conference on Computer Science & Education (ICCSE) · doi:10.1109/iccse.2019.8845325 · Cited by 34
“Neural Networks that have more than three layers and more than one hidden layer are called Deep Neural Networks [9].”
BackgroundOptimal Autonomous Driving Through Deep Imitation Learning and Neuroevolution
Jalali, S. M. J., Kebria, P. M., Khosravi, A., Saleh, K., Nahavandi, D., & Nahavandi, S. (2019) · Conference proceedings/Conference proceedings - IEEE International Conference on Systems, Man, and Cybernetics · doi:10.1109/smc.2019.8914582 · Cited by 34
“they obtained astonishing results in different applications, specially in the computer vision area [45].”
Are Accelerometers for Activity Recognition a Dead-end?
Tong, C., Tailor, S. A., & Lane, N. D. (2020) · IEEE Workshop on Mobile Computing Systems and Applications (WMCSA) · doi:10.1145/3376897.3377867 · Cited by 43
“overcome by substantial advancement seen in recent years: on one hand, powerful and efficient deep learning algorithms have been developed for image processing [14, 23, 32]; on the other hand, there”
BackgroundTernary Compression for Communication-Efficient Federated Learning
Xu, J., Du, W., Jin, Y., He, W., & Cheng, R. (2020) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2020.3041185 · Cited by 255
“2) Communication Costs : Federated learning is influenced by the rapidly growing depth of model and the amount of multiply-accumulate operations (MACs) [25].”
BackgroundA Learning Framework for n-Bit Quantized Neural Networks Toward FPGAs
Chen, J., Liu, L., Liu, Y., & Zeng, X. (2020) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2020.2980041 · Cited by 39
“NOTE THE GRAY RESULTS ARE IMPLEMENTED BY OUR n-BQ-NN, WHERE THE TRAINING METHOD OF 1-bit ACTIVATIONS IS INTRODUCED IN SECTION III-E, AND OTHER RESULTS ARE REPORTED BY [46].”
BackgroundPutra, R. V. W., Hanif, M. A., & Shafique, M. (2020) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1109/dac18072.2020.9218672 · Cited by 43
“, 100KB-500KB [13]), hence layer partitioning and scheduling are required.”
BackgroundEnergy Efficiency of Machine Learning in Embedded Systems Using Neuromorphic Hardware
Kang, M., Lee, Y.-S., & Park, M. (2020) · Electronics · doi:10.3390/electronics9071069 · Cited by 42
“When it is used as an accelerator in a system, other factors such as data transfer from the host system and control subsystem for interconnection should be considered because the energy cost of data movement is much higher than that of computation [16].”
MethodEfficiency Versus Accuracy: A Review of Design Techniques for DNN Hardware Accelerators
Latotzke, C., & Gemmeke, T. (2021) · IEEE Access · doi:10.1109/access.2021.3050670 · Cited by 35
“A classification of the data flow and its optimization are presented in [10], referring to systolic principles [12], and further in-depth analyses of the cost of a given hardware accelerator, i. e. energy and time-to-solution, were performed using an automation tool [13].”
MethodClassical and Deep Learning based Visual Servoing Systems: a Survey on State of the Art
Machkour, Z., Ortíz-Arroyo, D., & Durdevic, P. (2021) · Journal of Intelligent & Robotic Systems · doi:10.1007/s10846-021-01540-w · Cited by 40
“Several surveys have been published describing the recent advances in DNN’s architectures and its applications in computer vision tasks in [12, 98, 116, 137, 179, 207].”
BackgroundFilter Pruning via Learned Representation Median in the Frequency Domain
Zhang, X., Xie, W., Li, Y., Lei, J., & Du, Q. (2021) · IEEE Transactions on Cybernetics · doi:10.1109/tcyb.2021.3124284 · Cited by 35
“Reference [8] claims that VGGNet with 321.”
Efficient Hardware Acceleration of Sparsely Active Convolutional Spiking Neural Networks
Sommer, J., Özkan, M. A., Keszöcze, O., & Teich, J. (2022) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3197512 · Cited by 37
“Blue kernel: the input address event (0, 0)[5]in comes from sin = 5, which defines the center of the kernel.”
BackgroundMethodE-prop on SpiNNaker 2: Exploring online learning in spiking RNNs on neuromorphic hardware
Rostami, A., Vogginger, B., Yan, Y., & Mayr, C. (2022) · Frontiers in Neuroscience · doi:10.3389/fnins.2022.1018006 · Cited by 53
“By omitting the communication between edge and cloud, we can reduce power consumption of devices and increase their lifetimes (Sze et al., 2017; Chen and Ran, 2019; Cai et al., 2020).”
BackgroundProphet: Realizing a Predictable Real-time Perception Pipeline for Autonomous Vehicles
Liu, L., Dong, Z., Wang, Y., & Shi, W. (2022) · IEEE Real-Time Systems Symposium (RTSS) · doi:10.1109/rtss55097.2022.00034 · Cited by 34
“Reducing precision is usually achieved by reducing the number of bits/levels representing the data and reducing computation requirements and storage costs [44].”
BackgroundExplainable Methods for Image-Based Deep Learning: A Review
Gupta, L. K., Koundal, D., & Mongia, S. (2023) · Archives of Computational Methods in Engineering · doi:10.1007/s11831-023-09881-5 · Cited by 39
“In the last decade, the accuracy of deep learning methods has exceeded other machine learning techniques, and as a result, it is used in several real-world applications in areas such as computer vision, speech recognition, and robotics [1].”
BackgroundOptimized vision-directed deployment of UAVs for rapid traffic monitoring
Kyrkou, C., Timotheou, S., Kolios, P., Theocharides, T., & Panayiotou, C. G. (2018) · International Conference on Consumer Electronics · doi:10.1109/icce.2018.8326145 · Cited by 34
“, [14]) that will be able to run in real-time on the limited on-board hardware of UAVs.”
BackgroundSharma, V., Kim, H., & Kim, T. T.-H. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3168053 · Cited by 40
Dynamic Decision Tree Ensembles for Energy-Efficient Inference on IoT Edge Nodes
Daghero, F., Burrello, A., Macii, E., Montuschi, P., Poncino, M., & Pagliari, D. J. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3286276 · Cited by 50
“Nevertheless, the need for a constant connection to remote servers, especially in unstable or inse-cure environments which are common for IoT systems, may lead to unpredictable response latencies or confidentiality concerns [4], [5].”
BackgroundWei, W., Zhang, M., Qu, H., Belatreche, A., Zhang, J., & Chen, H. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.00968 · Cited by 37
Temiz, M., Zhang, Y., Fu, Y., Zhang, C., Meng, C., Kaplan, O., & Masouros, C. (2025) · IEEE Open Journal of the Communications Society · doi:10.1109/ojcoms.2025.3586560 · Cited by 51
“Lightweight DL models will also reduce the hardware cost and increase the energy efficiency.”
ReRAM-based accelerator for deep learning
Li, B., Song, L., Chen, F., Qian, X., Chen, Y., & Li, H. H. (2018) · Design, Automation, and Test in Europe · doi:10.23919/date.2018.8342118 · Cited by 33
Self-citationJin, H., Liu, B., Jiang, W., Ma, Y., Shi, X., He, B., & Zhao, S. (2018) · ACM Transactions on Architecture and Code Optimization · doi:10.1145/3243904 · Cited by 35
“To train a DNN model is to adjust its weight parameters to minimize the gap between the correct classes of the targets and the classes obtained by the trained DNN model [38].”
MethodSun, Y., Fei, T., Gao, S., & Pohl, N. (2019) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2019.8682277 · Cited by 34
“In this paper, we adopt a region-based deep convolutional neural network (R-DCNN) [9–11] to simultaneously detect and classify gestures, which are measured by a 77 GHz frequency-modulated continuous wave (FMCW) radar for invehicle infotainment and driver monitoring systems.”
MethodHuang, P. C., Li, Z., Dong, Z., Han, R., Zhou, Z., Zhu, D., Liu, L., Liu, X., & Kang, J. (2019) · ACS Applied Electronic Materials · doi:10.1021/acsaelm.9b00011 · Cited by 36
Ahmed, S., Lee, Y.-D., Hyun, S.-H., & Koo, I. (2019) · Energies · doi:10.3390/en12163091 · Cited by 36
“There are several types of normalization, such as min-max normalization, decimal scaling, and the standard deviation method [46,47].”
MethodStructured Pruning for Efficient Convolutional Neural Networks via Incremental Regularization
Wang, H., Hu, X., Zhang, Q., Wang, Y., Yu, L., & Hu, H. R. (2019) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2019.2961233 · Cited by 38
Jayakodi, N. K., Belakaria, S., Deshwal, A., & Doppa, J. R. (2020) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3366636 · Cited by 38
“Apart from the constraint of having an application-specific hardware, most of these solutions also have an overhead of analog-to-digital conversion [38].”
BackgroundClassification of diabetic retinopathy using unlabeled data and knowledge distillation
Abbasi, S., Hajabdollahi, M., Khadivi, P., Karimi, N., Roshandel, R., Shirani, S., & Samavi, S. (2021) · Artificial Intelligence in Medicine · doi:10.1016/j.artmed.2021.102176 · Cited by 37
Knapheide, J., Stabernack, B., & Kuhnke, M. (2020) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl50879.2020.00053 · Cited by 33
“An overview of various approaches to accelerating training and inference of CNNs is given in [1].”
A Survey Comparing Specialized Hardware And Evolution In TPUs For Neural Networks
Shahid, A., & Mushtaq, M. (2020) · IEEE International Multitopic Conference (INMIC) · doi:10.1109/inmic50486.2020.9318136 · Cited by 33
“To turn input to a correct output using mathematical manipulation Deep Neural Networks (DNN) are used, which consist of multi layers between the input and output layers [1].”
MethodTowards Reliable In-Memory Computing:From Emerging Devices to Post-von-Neumann Architectures
Amrouch, H., Du, N., Gebregiorgis, A., Hamdioui, S., & Polian, I. (2021) · IFIP/IEEE International Conference on Very Large Scale Integration · doi:10.1109/vlsi-soc53125.2021.9606966 · Cited by 37
Scheduling Massive Camera Streams to Optimize Large-Scale Live Video Analytics
Rong, C., Wang, J. H., Liu, J., Wang, J., Li, F., & Huang, X. (2021) · IEEE/ACM Transactions on Networking · doi:10.1109/tnet.2021.3125359 · Cited by 40
Energy-Efficient MAC Units for Fused Posit Arithmetic
Murillo, R., Mallasén, D., Barrio, A. A. D., & Botella, G. (2021) · IEEE International Conference on Computer Design: VLSI in Computers and Processors (ICCD) · doi:10.1109/iccd53106.2021.00032 · Cited by 33
“For this reason, MAC units are widely used in deep learning applications to perform and accelerate convolutions and matrix multiplications [10], [11].”
BackgroundAbbas, A., Baek, S., Silvera, N., Soulileuth, B., Pachepsky, Y. A., Ribolzi, O., Boithias, L., & Cho, K. H. (2021) · Hydrology and earth system sciences · doi:10.5194/hess-25-6185-2021 · Cited by 37
Xiao, T. P., Feinberg, B., Bennett, C. H., Agrawal, V. S., Saxena, P., Prabhakar, V., Ramkumar, K., Medu, H., Raghavan, V., Chettuvetty, R., et al. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2021.3134313 · Cited by 39
Classification of Machine Learning Algorithms
Taherdoost, H. (2022) · Lecture notes on data engineering and communications technologies · doi:10.1007/978-981-16-8403-6_38 · Cited by 34
Efficient Layer Compression Without Pruning
Wu, J., Zhu, D., Fang, L., Deng, Y. C., & Zhong, Z. (2023) · IEEE Transactions on Image Processing · doi:10.1109/tip.2023.3302519 · Cited by 37
“Recent works [7], [8], [9] have demonstrated that the over-parameterized DNNs have a large number of redundant weights which can be removed (i.e., set to zero).”
BackgroundWang, K., Dou, Y., Sun, T., Qiao, P., & Wen, D. (2022) · International Journal of Intelligent Systems · doi:10.1002/int.22883 · Cited by 37
LOPdM: A Low-Power On-Device Predictive Maintenance System Based on Self-Powered Sensing and TinyML
Chen, Z., Gao, Y., & Liang, J. (2023) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2023.3308251 · Cited by 55
“layer, which produces the final outputs of the network [46].”
BackgroundKim, Y., Hakak, S., & Ghorbani, A. A. (2024) · Computers & Security · doi:10.1016/j.cose.2024.103989 · Cited by 42
Energy-, Cost-, and Resource-Efficient IoT Hazard Detection System with Adaptive Monitoring
Kok, C. L., Heng, J. B., Koh, Y. Y., & Teo, T. H. (2025) · Sensors · doi:10.3390/s25061761 · Cited by 42
“The processing efficiency of deep neural networks, as outlined by Sze et al., underscores the importance of optimizing model architectures for embedded systems [45]. with an optimized balance of accuracy (F1 score: 85.9”
InfluentialOn How to Efficiently Implement Deep Learning Algorithms on PYNQ Platform
Stornaiuolo, L., Santambrogio, M. D., & Sciuto, D. (2018) · IEEE Computer Society Annual Symposium on VLSI · doi:10.1109/isvlsi.2018.00112 · Cited by 35
1.2 Intelligence on Silicon: From Deep-Neural-Network Accelerators to Brain Mimicking AI-SoCs
Yoo, H. (2019) · Digest of technical papers/Digest of technical papers - IEEE International Solid-State Circuits Conference · doi:10.1109/isscc.2019.8662469 · Cited by 39
“There have been many proposals for DNN accelerators [6] and many hardware architectures.”
BackgroundMethodInfluentialA comparative analysis of pooling strategies for convolutional neural network based Hindi ASR
Passricha, V., & Aggarwal, R. K. (2019) · Journal of Ambient Intelligence and Humanized Computing · doi:10.1007/s12652-019-01325-y · Cited by 35
Wang, D., Xu, K., Jia, Q., & Ghiasi, S. (2019) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3316781.3317753 · Cited by 32
“CNNs are composed of multiple functional layers [11], each of which performs certain type of arithmetic operations on the input image or feature map.”
BackgroundScaling Analysis of Specialized Tensor Processing Architectures for Deep Learning Models
Gordienko, Y. G., Kochura, Y., Taran, V., Gordienko, N., Rokovyi, A., Alienin, O., & Stirenko, S. (2019) · Studies in computational intelligence · doi:10.1007/978-3-030-31756-0_3 · Cited by 33
Hybrid neural network framework for detection of cyber attacks at smart infrastructures
Krundyshev, V., & Kalinin, M. (2019) · International Conference on Security of Information and Networks (SIN) · doi:10.1145/3357613.3357623 · Cited by 32
“V. Sze et al. in [19] examined deep neural networks (DNN), noting the advantages: it has a deep learning structure and the results of the work are of medium accuracy.”
Background
Metadata
- Authors
- Sze, V., Chen, Y.-H., Yang, T.-J., & Emer, J. S.
- Affiliations
- Massachusetts Institute of Technology (US)
- Nvidia (United States) (US)
- Venue
- Proceedings of the IEEE
- Published
- 2017-12
- Volume
- 105
- Issue
- 12
- Pages
- 2295-2329
- ISSN
- 0018-9219 (ISSN-L)
- 1558-2256
- OpenAlex
- W2604319603
- Cited by
- 4,215
- References
- 157
- Citations by year
- 2026: 504 · 2025: 523 · 2024: 474 · 2023: 466 · 2022: 508 · 2021: 568 · 2020: 551 · 2019: 392 · 2018: 199 · 2017: 28
- FWCI
- 132.75
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- Proceedings of the IEEE (publishedVersion)
- Funders
- DARPA YFA
- MIT CICS
- gift from Nvidia
- gift from Intel
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Memory and Neural Computing
- Keywords
- deep neural network, energy efficiency, throughput, hardware accelerators, computation cost reduction, benchmarking metrics
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.