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 301–400 of 1,011
Where Nanosensors Meet Machine Learning: Prospects and Challenges in Detecting Disease X
Leong, Y. X., Tan, E. X., Leong, S. X., Koh, C. S. L., Nguyen, L. B. T., Chen, J. R. T., Xia, K., & Ling, X. Y. (2022) · ACS Nano · doi:10.1021/acsnano.2c05731 · Cited by 88
Testability and Dependability of AI Hardware: Survey, Trends, Challenges, and Perspectives
Su, F., Liu, C., & Stratigopoulos, H. (2023) · IEEE Design and Test · doi:10.1109/mdat.2023.3241116 · Cited by 97
“Basic architectures include the streaming architecture composed of many cores with the layers mapped among the cores and the single-core architecture, that is, in the form of a systolic array that parallelizes the storage and computation of the different layers [3].”
BackgroundTransform Quantization for CNN Compression
Young, S. I., Wang, Z., Taubman, D., & Girod, B. (2021) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2021.3084839 · Cited by 109
“CNN weights no longer fit in the cache at the same time, and they must be loaded from the off-chip memory as required, which can be very energy-consuming [6].”
BackgroundEfficient Multiple-Precision Floating-Point Fused Multiply-Add with Mixed-Precision Support
Zhang, H., Chen, D., & Ko, S. (2019) · IEEE Transactions on Computers · doi:10.1109/tc.2019.2895031 · Cited by 77
“Use of half-precision is suggested in deep learning applications to take the place of the conventional 32-bit floating-point format [16], [17].”
MethodRapuano, E., Meoni, G., Pacini, T., Dinelli, G., Furano, G., Giuffrida, G., & Fanucci, L. (2021) · Remote Sensing · doi:10.3390/rs13081518 · Cited by 96
“The main advantage is the reduced number of needed access to the external memory (only the input image shall be read), drastically reducing latency and power consumption [34,35].”
BackgroundLee, E., Han, T., Seo, D., Shin, G., Kim, J., Kim, S., Jeong, S., Rhe, J., Park, J., Ko, J. H., et al. (2021) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2021.3080042 · Cited by 75
Park, H.-S., & Park, D. Y. (2021) · Building and Environment · doi:10.1016/j.buildenv.2021.108492 · Cited by 79
FPGA-based Deep Learning Inference Accelerators: Where Are We Standing?
Nechi, A., Groth, L., Mulhem, S., Merchant, F., Buchty, R., & Bereković, M. (2023) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3613963 · Cited by 111
“This increase in DNN model size led to a remarkable development of DNN architectures [144] and increased accuracy, albeit requiring more computing and memory resources.”
BackgroundSanity-Check: Boosting the Reliability of Safety-Critical Deep Neural Network Applications
Ozen, E., & Orailoğlu, A. (2019) · Proceedings - Asian Test Symposium/Proceedings · doi:10.1109/ats47505.2019.000-8 · Cited by 82
“Let us consider the convolution layer formula with the notation introduced in [14]:”
BackgroundNguyen, T., Phan, T. T. V., Ho, D.-D., Pradhan, A. M. S., & Huynh, T. (2022) · Engineering Structures · doi:10.1016/j.engstruct.2022.114172 · Cited by 81
Single-shot optical neural network
Bernstein, L., Sludds, A., Panuski, C., Trajtenberg‐Mills, S., Hamerly, R., & Englund, D. R. (2023) · Science Advances · doi:10.1126/sciadv.adg7904 · Cited by 100
Moving Deep Learning to the Edge
Véstias, M. P., Duarte, R. P., de Sousa, J. T., & Neto, H. C. (2020) · Algorithms · doi:10.3390/a13050125 · Cited by 87
“On the edge, the design of DNNs is also subject to energy, cost, and latency constraints.”
BackgroundInfluentialFruit Classification for Retail Stores Using Deep Learning
Rojas-Aranda, J. L., Nunez-Varela, J. I., Cuevas-Tello, J. C., & Rangel-Ramirez, G. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-49076-8_1 · Cited by 94
Rybalkin, V., Pappalardo, A., Ghaffar, M. M., Gambardella, G., Wehn, N., & Blott, M. (2018) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl.2018.00024 · Cited by 92
Deep learning in target prediction and drug repositioning: Recent advances and challenges
Yu, J., Dai, Q.-Q., & Li, G. (2021) · Drug Discovery Today · doi:10.1016/j.drudis.2021.10.010 · Cited by 82
A Scalable Near-Memory Architecture for Training Deep Neural Networks on Large In-Memory Datasets
Schuiki, F., Schaffner, M., Gürkaynak, F. K., & Benini, L. (2018) · IEEE Transactions on Computers · doi:10.1109/tc.2018.2876312 · Cited by 86
In-memory computing with emerging nonvolatile memory devices
Cheng, C., Tiw, P. J., Cai, Y., Yan, X., Yang, Y., & Huang, R. (2021) · Science China Information Sciences · doi:10.1007/s11432-021-3327-7 · Cited by 77
Rangel, G., Cuevas-Tello, J. C., Nunez-Varela, J. I., Puente, C., & Trujillo, A. G. S. (2024) · Journal of Sensors · doi:10.1155/2024/2797320 · Cited by 112
Kim, S., & Kim, H. (2021) · IEEE Access · doi:10.1109/access.2021.3054879 · Cited by 71
“In HW architectures, floating-point arithmetic requires much more energy consumption for addition and multiplication operations than fixed-point arithmetic [24].”
BackgroundMethodInfluentialLocalization and recognition of leukocytes in peripheral blood: A deep learning approach
Reena, M. R., & Ameer, P. M. (2020) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2020.104034 · Cited by 79
Libraries of Approximate Circuits: Automated Design and Application in CNN Accelerators
Mrázek, V., Sekanina, L., & Vašíček, Z. (2020) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2020.3032495 · Cited by 71
“Various fixed-point as well as floating-point number representations were evaluated for CNNs to reduce power consumption [10], [33].”
BackgroundMethodInfluentialSegmentation of leukocyte by semantic segmentation model: A deep learning approach
Roy, R. M., & Ameer, P. M. (2020) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2020.102385 · Cited by 82
Viale, A., Marchisio, A., Martina, M., Masera, G., & Shafique, M. (2021) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn52387.2021.9533738 · Cited by 94
“The main methods to make these decisions in fast and accurate ways are represented by Deep Learning algorithms, that are typically divided into the following types: • Deep Neural Networks (DNNs), that are the oldest and can be implemented on traditional hardware processors and specialized architectures [6][7].”
MethodRETRACTED ARTICLE: A review on quantum computing and deep learning algorithms and their applications
Valdez, F., & Melín, P. (2022) · Soft Computing · doi:10.1007/s00500-022-07037-4 · Cited by 92
RetractedAn Interpretable Neural Network for Configuring Programmable Wireless Environments
Liaskos, C., Tsioliaridou, A., Nie, S., Pitsillides, A., Ioannidis, S., & Akyildiz, I. F. (2019) · IEEE International Workshop on Signal Processing Advances in Wireless Communications (SPAWC) · doi:10.1109/spawc.2019.8815428 · Cited by 65
“A neural network consists of an input layer with input units, at least one layer of hidden nodes neurons, and an output layer with multiple output units [14].”
BackgroundRes-DNN: A Residue Number System-Based DNN Accelerator Unit
Samimi, N., Kamal, M., Afzali‐Kusha, A., & Pedram, M. (2019) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2019.2951083 · Cited by 80
“tend to have more layers to achieve higher accuracy [21].”
BackgroundDual integrated convolutional neural network for real-time facial expression recognition in the wild
Saurav, S., Gidde, P., Saini, R., & Singh, S. K. (2021) · The Visual Computer · doi:10.1007/s00371-021-02069-7 · Cited by 73
Tripathi, N., Goshisht, M. K., Sahu, S. K., & Arora, C. (2021) · Molecular Diversity · doi:10.1007/s11030-021-10237-z · Cited by 80
Hou, H., Jawaddi, S. N. A., & Ismail, A. B. (2023) · Future Generation Computer Systems · doi:10.1016/j.future.2023.10.002 · Cited by 78
Memory-Efficient Deep Learning on a SpiNNaker 2 Prototype
Liu, C., Bellec, G., Vogginger, B., Kappel, D., Partzsch, J., Neumärker, F., Höppner, S., Maass, W., Furber, S. B., Legenstein, R. A., et al. (2018) · Frontiers in Neuroscience · doi:10.3389/fnins.2018.00840 · Cited by 86
“, for medical applications, robotics, or tactile devices, their physical realization needs to be extremely energy efficient to meet requirements on durability, size, and heat emission (König et al., 2002; Sze et al., 2017).”
BackgroundParticle Filter Networks with Application to Visual LocalizationPreprint
Karkus, P., Hsu, D., & Lee, W. S. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1805.08975 · Cited by 162
“We may also improve inference time, leveraging an abundance of work optimizing neural network models and hardware [45].”
MethodLDFR: Learning deep feature representation for software defect prediction
Xu, Z., Li, S., Xu, J., Liu, J., Luo, X., Zhang, Y., Zhang, T., Keung, J., & Tang, Y. (2019) · Journal of Systems and Software · doi:10.1016/j.jss.2019.110402 · Cited by 75
Large graph convolutional network training with GPU-oriented data communication architecture
Min, S., Wu, K., Huang, S., Hidayetoğlu, M., Xiong, J., Ebrahimi, E., Chen, D., & Hwu, W. (2021) · Proceedings of the VLDB Endowment · doi:10.14778/3476249.3476264 · Cited by 98
“Acceleration of modern machine learning models is often severely limited by insufficient memory bandwidth [24, 42, 43].”
BackgroundTrends in human activity recognition with focus on machine learning and power requirements
Nguyen, B., Coelho, Y., Bastos-Filho, T., & Krishnan, S. (2021) · Machine Learning with Applications · doi:10.1016/j.mlwa.2021.100072 · Cited by 93
Analog Nanoscale Electro-Optical Synapses for Neuromorphic Computing Applications
Portner, K., Schmuck, M., Lehmann, P., Weilenmann, C., Haffner, C., Ma, P., Leuthold, J., Luisier, M., & Emboras, A. (2021) · ACS Nano · doi:10.1021/acsnano.1c04654 · Cited by 71
Emerging memristive neurons for neuromorphic computing and sensing
Li, Z., Tang, W., Zhang, B., Yang, R., & Miao, X. (2023) · Science and Technology of Advanced Materials · doi:10.1080/14686996.2023.2188878 · Cited by 76
“ANNs can perform supervised, semi-supervised, unsupervised, and reinforcement learning algorithms, and excel in deep learning tasks that had large amounts of computational resources for the training data [2,122].”
MethodGreen Federated Learning: A New Era of Green Aware AI
Thakur, D., Guzzo, A., Fortino, G., & Piccialli, F. (2025) · ACM Computing Surveys · doi:10.1145/3718363 · Cited by 108
“Additionally, it dynamically adjusts the buffer size used in asynchronous aggregation to optimize the trade-off between staleness and runtime.”
Entropy-Aware I/O Pipelining for Large-Scale Deep Learning on HPC Systems
Zhu, Y., Chowdhury, F., Fu, H., Moody, A. T., Mohror, K., Sato, K., & Yu, W. (2018) · Proceedings - International Symposium on Modeling, Analysis, and Simulation of Computer and Telecommunication Systems · doi:10.1109/mascots.2018.00023 · Cited by 85
“g, GoogleNet [44], ResNet [26]) with larger filters and more weights are adopting deeper layerswith larger datasets (LabelMe [40], ImageNet [19]), because large datasets can prevent deep neural network (DNN) from overfitting during training by utilizing more iterations to update an increased number of weight parameter…”
Deep learning and control algorithms of direct perception for autonomous driving
Lee, D., Chen, K.-L., Liou, K.-H., Liu, C.-L., & Liu, J. (2020) · Applied Intelligence · doi:10.1007/s10489-020-01827-9 · Cited by 73
Affective design using machine learning: a survey and its prospect of conjoining big data
Chan, K. Y., Kwong, C. K., Wongthongtham, P., Jiang, H., Fung, C. K. Y., Abu-Salih, B., Liu, Z., Wong, T. C., & Jain, P. (2018) · International Journal of Computer Integrated Manufacturing · doi:10.1080/0951192x.2018.1526412 · Cited by 68
Analog/Mixed-Signal Hardware Error Modeling for Deep Learning Inference
Rekhi, A. S., Zimmer, B., Nedovic, N., Liu, N., Venkatesan, R., Wang, M., Khailany, B., Dally, W. J., & Gray, C. T. (2019) · Proceedings of the 56th Annual Design Automation Conference 2019 · doi:10.1145/3316781.3317770 · Cited by 67
“that the reuse of weights and/or activations significantly reduces the energy cost of memory access [5].”
BackgroundA Survey on the Optimization of Neural Network Accelerators for Micro-AI On-Device Inference
Mazumder, A. N., Meng, J., Rashid, H.-A., Kallakuri, U., Zhang, X., Seo, J.-S., & Mohsenin, T. (2021) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2021.3129415 · Cited by 88
“Additionally, authors in [1] and [105] also provide different data-flow schemes that exploit the aforementioned data reuse opportunities, In weight stationary data-flow, the aim is to reuse the filter values read from the larger memory.”
BackgroundMethodInfluentialEstimation of Greenhouse Lettuce Growth Indices Based on a Two-Stage CNN Using RGB-D Images
Gang, M.-S., Kim, H.-J., & Kim, D. (2022) · Sensors · doi:10.3390/s22155499 · Cited by 73
“Deep learning techniques do not require handcrafted features selected by experts and provide a relatively fast processing speed in inference compared to traditional computer vision-based methods [11].”
BackgroundAmirkhani, A., Karimi, M. P., & Banitalebi-Dehkordi, A. (2022) · The Visual Computer · doi:10.1007/s00371-022-02660-6 · Cited by 82
“…improvement of the deep learning models [10], discovering the importance of the different deep learning applications [11], hardware advancements [12], the increased power of the computer processors and the graphical processing units [13], and the development of various software libraries and…”
BackgroundVijayakumar, K. P., Pradeep, K., Balasundaram, A., & Prusty, M. R. (2023) · Processes · doi:10.3390/pr11041072 · Cited by 66
Small-Footprint Keyword Spotting on Raw Audio Data with Sinc-Convolutions
Mittermaier, S., Kürzinger, L., Waschneck, B., & Rigoll, G. (2020) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp40776.2020.9053395 · Cited by 84
Sun, J., Zhang, X., & Wang, J. (2022) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2022.105662 · Cited by 71
Madhushani, C. I., Dananjaya, K. G. S., Ekanayake, I. U., Meddage, D. P. P., Kantamaneni, K., & Rathnayake, U. (2024) · Journal of Hydrology · doi:10.1016/j.jhydrol.2024.130846 · Cited by 74
Higher SNR PET image prediction using a deep learning model and MRI image
Liu, C., & Qi, J. (2019) · Physics in Medicine and Biology · doi:10.1088/1361-6560/ab0dc0 · Cited by 69
“Deep learning (DL) has been reemerging recently in many fields, including computer vision and speech-recognition, because of big data and groundbreaking GPU performance (LeCun et al 2015, Sze et al 2017).”
BackgroundEnergy-Efficient Processing and Robust Wireless Cooperative Transmission for Edge Inference
Yang, K., Shi, Y., Yu, W., & Ding, Z. (2020) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.2979523 · Cited by 65
“In addition to the low-latency, improving the energy efficiency [5] is also critical due to the high computational complexity of processing DNNs, for which a number of works focusing on model compression methods [6], [7].”
Deep learning approach to detect seizure using reconstructed phase space images
Ilakiyaselvan, N., Khan, A. N., & Shahina, A. (2020) · Journal of Biomedical Research · doi:10.7555/jbr.34.20190043 · Cited by 71
Gong, M., Pan, K., Xie, Y., Qin, A. K., & Tang, Z. (2020) · Neural Networks · doi:10.1016/j.neunet.2020.02.001 · Cited by 74
Predicting Temperature of Permanent Magnet Synchronous Motor Based on Deep Neural Network
Guo, H., Ding, Q., Song, Y., Tang, H., Wang, L., & Zhao, J. (2020) · Energies · doi:10.3390/en13184782 · Cited by 64
Jeon, W., Ko, G., Lee, J., Lee, H., Ha, D., & Ro, W. W. (2021) · Advances in computers · doi:10.1016/bs.adcom.2020.11.003 · Cited by 84
Environment-Aware and Training-Free Beam Alignment for mmWave Massive MIMO via Channel Knowledge Map
Wu, D., Zeng, Y., Jin, S., & Zhang, R. (2021) · IEEE International Conference on Communications workshops · doi:10.1109/iccworkshops50388.2021.9473871 · Cited by 64
“Based on such finite data points collected, various machine learning techniques such as Knearest neighbor (KNN) [22], Kridging-based method [23], or Deep Neural Networks (DNNs) [24] could be applied to train the entire CPM or BIM.”
MethodEfficient Posit Multiply-Accumulate Unit Generator for Deep Learning Applications
Zhang, H., He, J., & Ko, S. (2019) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas.2019.8702349 · Cited by 67
“Due to these advantages, in deep learning applications, MAC units are widely used to perform convolution or dot-product operations [8] [9] [10].”
BackgroundDemetriou, D., Michailides, C., Papanastasiou, G., & Onoufriou, T. (2021) · Ocean Engineering · doi:10.1016/j.oceaneng.2021.108592 · Cited by 63
EdgeDuet: Tiling Small Object Detection for Edge Assisted Autonomous Mobile Vision
Yang, Z., Wang, X., Wu, J., Zhao, Y., Ma, Q., Miao, X., Zhang, L., & Zhou, Z. (2022) · IEEE/ACM Transactions on Networking · doi:10.1109/tnet.2022.3223412 · Cited by 82
“Model compression techniques can dramatically reduce the workload of deep learning models [9].”
BackgroundPhotonic Neural Networks Based on Integrated Silicon Microresonators
Biasi, S., Donati, G. P., Lugnan, A., Mancinelli, M., Staffoli, E., & Pavesi, L. (2023) · Intelligent Computing · doi:10.34133/icomputing.0067 · Cited by 79
“ANNs are usually implemented on electronic hardware which are based on Von Neuman architectures, such as general purpose CPUs (Central Processing Units), massively parallel GPUs (Graphical Processing Units), and specialized integrated circuits dedicated to accelerate specific operations like TPUs (Tensor Processing Un…”
MethodSELD-TCN: Sound Event Localization & Detection via Temporal Convolutional Networks
Guirguis, K., Schorn, C., Guntoro, A., Abdulatif, S., & Yang, B. (2020) · European Signal Processing Conference (EUSIPCO) · doi:10.23919/eusipco47968.2020.9287716 · Cited by 63
“However, the advances in hardware accelerators enable the realization of such architectures while reducing the energy cost, silicon area, and memory size requirements [38].”
BackgroundEyeriss v2: A Flexible and High-Performance Accelerator for Emerging Deep Neural NetworksPreprint
Chen, Y., Emer, J., & Sze, V. (2018) · arXiv (Cornell University) · Cited by 63
Self-citationImpact of Approximate Multipliers on VGG Deep Learning Network
Hammad, I., & El‐Sankary, K. (2018) · IEEE Access · doi:10.1109/access.2018.2875376 · Cited by 69
“According to [12], the VGGNet-16 required 15.”
(2018) · International Conference on Learning Representations · Cited by 62
Knowledge Transfer via Dense Cross-Layer Mutual-Distillation
Yao, A., & Sun, D. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58555-6_18 · Cited by 78
FlexDNN: Input-Adaptive On-Device Deep Learning for Efficient Mobile Vision
Fang, B., Zeng, X., Zhang, F., Xu, H. F., & Zhang, M. (2020) · IEEE/ACM Symposium on Edge Computing (SEC) · doi:10.1109/sec50012.2020.00014 · Cited by 90
“Unfortunately, DNNs are known to be computation-expensive [31], and high computation cost directly translates to high processing latency and energy consumption.”
BackgroundYadav, R. K., Raj, S., & Pradhan, A. K. (2019) · IEEE Transactions on Smart Grid · doi:10.1109/tsg.2019.2912350 · Cited by 62
“However, rectified linear unit (reLU) and exponential linear unit (eLU) activation functions enable fast network training with less complexity and introduces sparsity in the network, thus preferred for DNN training in method [27].”
BackgroundMethodInfluentialAn Efficient Bit-Flip Resilience Optimization Method for Deep Neural Networks
Schorn, C., Guntoro, A., & Ascheid, G. (2019) · Design, Automation, and Test in Europe · doi:10.23919/date.2019.8714885 · Cited by 67
“However, when these networks are deployed on a dedicated accelerator, bit-width can be reduced to save energy [2].”
BackgroundMethodDeep compressive autoencoder for action potential compression in large-scale neural recording
Wu, T., Zhao, W., Keefer, E., & Yang, Z. (2018) · Journal of Neural Engineering · doi:10.1088/1741-2552/aae18d · Cited by 66
“3G multiply-accumulates (MAC) to process one image [47].”
BackgroundCastro‐Zunti, R. D., Park, E. H., Choi, Y., Jin, G. Y., & Ko, S. (2020) · Computerized Medical Imaging and Graphics · doi:10.1016/j.compmedimag.2020.101718 · Cited by 67
The Hardware and Algorithm Co-Design for Energy-Efficient DNN Processor on Edge/Mobile Devices
Lee, J., Kang, S., Lee, J., Shin, D., Han, D., & Yoo, H. (2020) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2020.3021397 · Cited by 68
“Compared to the previous review papers [9]–[11], this paper analysis the latest mobile/edge DNN accelerators and discusses how”
ResultAttention-Based Deep Entropy Active Learning Using Lexical Algorithm for Mental Health Treatment
Ahmed, U., Mukhiya, S. K., Srivastava, G., Lamo, Y., & Lin, J. C. (2021) · Frontiers in Psychology · doi:10.3389/fpsyg.2021.642347 · Cited by 63
“The major difference between architectures is the hidden layers, layers type, shapes, and connection between layers (Sze et al., 2017).”
BackgroundPower Side-Channel Attacks on BNN Accelerators in Remote FPGAs
Moini, S., Tian, S., Holcomb, D. E., Szefer, J. M., & Tessier, R. (2021) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2021.3074608 · Cited by 88
“Deep neural networks (DNNs) [31] are a class of artificial neural networks that use multiple layers.”
BackgroundDual-stage attention-based LSTM for simulating performance of brackish water treatment plant
Yoon, N., Kim, J., Lim, J., Abbas, A., Jeong, K., & Cho, K. H. (2021) · Desalination · doi:10.1016/j.desal.2021.115107 · Cited by 64
Towards Mixed-Precision Quantization of Neural Networks via Constrained Optimization
Chen, W., Wang, P., & Cheng, J. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00530 · Cited by 104
“Here we only review the works related to ours and refer the reader to recent surveys [12, 6, 32, 5, 8] for a comprehensive overview.”
BackgroundToward memristive in-memory computing: principles and applications
Bao, H., Zhou, H., Li, J., Pei, H.-Z., Tian, J., Yang, L., Ren, S., Tong, S.-Q., Li, Y., He, Y., et al. (2022) · Frontiers of Optoelectronics · doi:10.1007/s12200-022-00025-4 · Cited by 79
“Basic image edge detection methods usually use convolutional operations to find the locations where the gray value changes abruptly, which are similar to those in the CNNs, except that the kernels are pre-defined rather than being trained.”
BackgroundInfluentialPan, Y., Fang, W., & Zhang, W. (2023) · Journal of Cleaner Production · doi:10.1016/j.jclepro.2023.136742 · Cited by 77
Shahcheraghian, A., Madani, H., & Ilinca, A. (2024) · Energies · doi:10.3390/en17020376 · Cited by 61
“Despite these hurdles, the substantial predictive power of DNNs and ongoing research in model interpretability ensure their continued prominence in the future of the energy industry, where data-driven decision-making and sustainability are paramount [111,116].”
BackgroundVision-Language Models for Edge Networks: A Comprehensive Survey
Sharshar, A., Khan, L. U., Ullah, W., & Guizani, M. (2025) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3579032 · Cited by 125
“Moreover, purpose-built hardware accelerators (e.g., Google’s Edge TPU) and edge-native architectures have also played a crucial role in enhancing the accessibility in deploying VLMs on edge-constrained hardware [12], [13].”
Race Recognition Using Deep Convolutional Neural Networks
Vo, T., Nguyen, T., & Le, C. T. (2018) · Symmetry · doi:10.3390/sym10110564 · Cited by 59
Kuang, Y., Cui, X., Zhong, Y., Liu, K., Zou, C., Dai, Z., Wang, Y., Yu, D., & Huang, R. (2021) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2021.3052172 · Cited by 60
An End-to-End Anti-Jamming Target Detection Method Based on CNN
Zhang, Y., Jiu, B., Wang, P., Liu, H., & Liang, S. (2021) · IEEE Sensors Journal · doi:10.1109/jsen.2021.3103042 · Cited by 60
A 95.6-TOPS/W Deep Learning Inference Accelerator With Per-Vector Scaled 4-bit Quantization in 5 nm
Keller, B., Venkatesan, R., Dai, S. X., Tell, S. G., Zimmer, B., Sakr, C., Dally, W. J., Gray, C. T., & Khailany, B. (2023) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2023.3234893 · Cited by 74
“While some prior works claim that memory accesses dominate DL inference accelerator energy costs [52], [53], [54], [55], we demonstrate that an efficient implementation can greatly reduce non-datapath overheads.”
ResultACE-CNN: Approximate Carry Disregard Multipliers for Energy-Efficient CNN-Based Image Classification
Shakibhamedan, S., Amirafshar, N., Baroughi, A. S., Shahhoseini, H. S., & TaheriNejad, N. (2024) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2024.3369230 · Cited by 65
“Fundamentally, the most computationally intensive operation is the multiplication operation for convolutional layers [9].”
BackgroundOh, C., Jo, M., & Son, J. (2019) · ACS Applied Materials & Interfaces · doi:10.1021/acsami.9b00392 · Cited by 63
Photometry of high-redshift blended galaxies using deep learning
Boucaud, A., Huertas-Company, M., Heneka, C., Ishida, E. E. O., Sedaghat, N., de Souza, R. S., Moews, B., Dole, H., Castellano, M., Merlin, E., et al. (2020) · Monthly Notices of the Royal Astronomical Society · doi:10.1093/mnras/stz3056 · Cited by 60
“Sze et al. 2017) to directly compute the fluxes of the two galaxies from the blend image.”
BackgroundPerformance Evaluation of Deep Learning Tools in Docker Containers
Xu, P., Shi, S., & Chu, X. (2017) · International Conference on Big Data Computing and Communications (BigCom) · doi:10.1109/bigcom.2017.32 · Cited by 75
Opportunities for integrated photonic neural networks
Stark, P., Horst, F., Dangel, R., Weiß, J., & Offrein, B. J. (2018) · Nanophotonics · doi:10.1515/nanoph-2020-0297 · Cited by 70
Chen, B., You, S., Yang, Y., Fu, Y., Ye, Z., Deng, J., Wang, K., & Hong, Y. E. (2021) · The Science of The Total Environment · doi:10.1016/j.scitotenv.2020.144724 · Cited by 61
Visual explanations from spiking neural networks using inter-spike intervals
Kim, Y., & Panda, P. (2021) · Scientific Reports · doi:10.1038/s41598-021-98448-0 · Cited by 91
An Xception Based Convolutional Neural Network for Scene Image Classification with Transfer Learning
Wu, X., Liu, R., Yang, H., & Chen, Z. (2020) · International Conference on Information Technology and Computer Application (ITCA) · doi:10.1109/itca52113.2020.00063 · Cited by 65
“The Inception structure is similar to using a 1x1 convolution layer to learn the association of features between channels from the input feature map, and then segment the output feature map and process the following 3x3 convolution layer to deal with the association of spatial elements [10].”
Methoddel Hougne, P., & Lerosey, G. (2018) · Physical Review X · doi:10.1103/physrevx.8.041037 · Cited by 92
“4% on standard tasks [75]; moreover, if the ANN has been specifically designed for low resolution, it can achieve similar results with as little as 1-bit or 2-bit resolution.”
Hybrid machine learning assisted modelling framework for particle processes
Nielsen, R. F., Nazemzadeh, N., Sillesen, L. W., Andersson, M., Gernaey, K. V., & Mansouri, S. S. (2020) · Computers & Chemical Engineering · doi:10.1016/j.compchemeng.2020.106916 · Cited by 68
Big Data Analytics for Large Scale Wireless Networks: Challenges and OpportunitiesPreprint
Dai, H., Wong, R. C.-W., Wang, H. H., Zheng, Z., & Vasilakos, A. V. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1909.08069 · Cited by 60
“Recently, deep learning architectures have received extensive attention [190].”
BackgroundTowards Energy-Efficient and Secure Edge AI: A Cross-Layer Framework ICCAD Special Session Paper
Shafique, M., Marchisio, A., Wicaksana Putra, R. V., & Hanif, M. A. (2021) · 2021 IEEE/ACM International Conference On Computer Aided Design (ICCAD) · doi:10.1109/iccad51958.2021.9643539 · Cited by 58
“Therefore, redundant accesses for the same data to DRAM are required, which restricts the systems from achieving high performance and energy efficiency gains, as DRAM access latency and energy are significantly higher than other operations [32].”
BackgroundEdgeDuet: Tiling Small Object Detection for Edge Assisted Autonomous Mobile Vision
Wang, X., Yang, Z., Wu, J., Zhao, Y., & Zhou, Z. (2021) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom42981.2021.9488843 · Cited by 72
“Model compression techniques can dramatically reduce the workload of deep learning models [5].”
BackgroundA Lightweight Framework for Human Activity Recognition on Wearable Devices
Coelho, Y., dos Santos, F. D. A. S., Frizera, A., & Bastos-Filho, T. (2021) · IEEE Sensors Journal · doi:10.1109/jsen.2021.3113908 · Cited by 65
Digital twin-driven clamping force control for thin-walled parts
Wang, G., Cao, Y., & Zhang, Y. (2021) · Advanced Engineering Informatics · doi:10.1016/j.aei.2021.101468 · Cited by 61
DL4SciVis: A State-of-the-Art Survey on Deep Learning for Scientific Visualization
Wang, C., & Han, J. (2022) · IEEE Transactions on Visualization and Computer Graphics · doi:10.1109/tvcg.2022.3167896 · Cited by 77
“With the impressive advances in graphics hardware [28], [118], [138] and DL architectures (e.g., AlexNet [90] for image classification, GAN [42] for image generation, and U-Net [122] for image segmentation), the renaissance of artificial intelligence (AI) as a viable solution for solving challenging…”
BackgroundDesigning Hardware for Machine Learning: The Important Role Played by Circuit Designers
Sze, V. (2017) · IEEE Solid-State Circuits Magazine · doi:10.1109/mssc.2017.2745798 · Cited by 64
“A more detailed discussion on DNNs can be found in [17].”
BackgroundSelf-citationSim, J., Lee, S., & Kim, L. (2019) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2019.2935251 · Cited by 62
“of multiply–accumulate (MAC) operations and ten millions to hundred millions of weights to process a single image, limiting its usage in energy-poor mobile environments [6].”
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.