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 501–600 of 1,011
An Experimental Analysis of Deep Learning Architectures for Supervised Speech Enhancement
Nossier, S. A., Wall, J. A., Moniri, M., Glackin, C., & Cannings, N. (2020) · Electronics · doi:10.3390/electronics10010017 · Cited by 49
FPGA-based implementation of classification techniques: A survey
Saidi, A., Othman, S. B., Dhouibi, M., & Saoud, S. B. (2021) · Integration · doi:10.1016/j.vlsi.2021.08.004 · Cited by 55
Fast and robust analog in-memory deep neural network training
Rasch, M. J., Carta, F., Fagbohungbe, O. I., & Gokmen, T. (2024) · Nature Communications · doi:10.1038/s41467-024-51221-z · Cited by 74
“In-memory computing with resistive crossbar arrays could greatly accelerate deep-learning workloads, since energy efficient acceleration can be achieved by implementing ubiquitous matrix-vector multiplications (MVMs) using resistive elements and fundamental physics (Kirchhoff’s and Ohm’s laws) (Sebastian et al., 2020;…”
Culturomics: Taking the cross-scale, interdisciplinary science of culture into the next decade
Luo, S., Yuan, H., Wang, Y., & Bond, M. H. (2024) · Neuroscience & Biobehavioral Reviews · doi:10.1016/j.neubiorev.2024.105942 · Cited by 50
A new deep learning method based on AlexNet model and SSD model for tennis ball recognition
Gu, S., Ding, L., Yang, Y., & Chen, X. (2017) · IEEE International Workshop on Computational Intelligence and Applications (IWCIA) · doi:10.1109/iwcia.2017.8203578 · Cited by 45
“In 2012, at the ImageNet challenge, a group from the University of Toronto used graphics processing units (GPU) for their high compute capability and a deep neural network approach, namely AlexNet, and dropped the error rate of the algorithms by approximately 10 % [7].”
MethodDeep neural network–based soft computing the resonant frequency of E–shaped patch antennas
Üstün, D., Toktaş, A., & Akdağlı, A. (2019) · AEU - International Journal of Electronics and Communications · doi:10.1016/j.aeue.2019.02.011 · Cited by 45
Exploiting time-frequency patterns with LSTM-RNNs for low-bitrate audio restoration
Deng, J., Schuller, B. W., Eyben, F., Schuller, D., Zhang, Z., Francois, H., & Oh, E. (2019) · Neural Computing and Applications · doi:10.1007/s00521-019-04158-0 · Cited by 49
“Thanks to their capability of extracting representative features, CNNs have been widely and successfully applied in image classification [58].”
Hanif, M. A., & Shafique, M. (2019) · Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences · doi:10.1098/rsta.2019.0164 · Cited by 46
Ten quick tips for deep learning in biology
Lee, B. D., Gitter, A., Greene, C. S., Raschka, S., Maguire, F., Titus, A., Kessler, M. D., Lee, A. J., Chevrette, M. G., Stewart, P. A., et al. (2022) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1009803 · Cited by 50
“Such training can be made more di cult by the demanding nature of these deep networks, which often require extensive time investments into tuning and computing infrastructure to achieve state-of-the-art performance [18].”
BackgroundRadarSNN: A Resource Efficient Gesture Sensing System Based on mm-Wave Radar
Arsalan, M., Santra, A., & Іссаков, В. (2022) · IEEE Transactions on Microwave Theory and Techniques · doi:10.1109/tmtt.2022.3148403 · Cited by 50
“The majority of the energy is consumed by the multiply–accumulate (MAC) operations between layers [43].”
BackgroundGroup contribution-based property modeling for chemical product design: A perspective in the AI era
Mann, V., Gani, R., & Venkatasubramanian, V. (2023) · Fluid Phase Equilibria · doi:10.1016/j.fluid.2023.113734 · Cited by 46
The Role of Artificial Intelligence and Machine Learning Techniques: Race for COVID-19 Vaccine
Kannan, S., Subbaram, K., Ali, S., & Kannan, H. (2020) · Archives of Clinical Infectious Diseases · doi:10.5812/archcid.103232 · Cited by 53
“Although its impact may seem relatively limited at first, this technology has great potential in the field of drug discovery and development (11).”
BackgroundStephen, O., Maduh, U. J., & Sain, M. (2021) · Electronics · doi:10.3390/electronics11010055 · Cited by 51
Ravikumar, A., Sriraman, H., Saketh, P. M. S., Lokesh, S., & Karanam, A. (2022) · PeerJ Computer Science · doi:10.7717/peerj-cs.909 · Cited by 48
“The input from the previous layers is fed to the fully connected layer, which gives the classification output based on the previous layer features extracted (Sze et al., 2017).”
BackgroundMethodMan, C. K., Quddus, M., Theofilatos, A., Yu, R., & Imprialou, M. (2022) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2022.3207798 · Cited by 51
“MLP is a subset of Deep Neural Network under the condition which more than three fully connected layers are presented in the MLP architecture [57].”
BackgroundZheng, L., Long, M., Chen, B., & Fan, Y. (2023) · International Journal of Educational Technology in Higher Education · doi:10.1186/s41239-023-00415-4 · Cited by 70
“As a deep learning technique, DNNs have demonstrated superior abilities in extracting high-level features (Sze et al., 2017).”
MethodA deep learning approach towards autonomous flight in forest environments
Dionisio-Ortega, S., Rojas-Perez, L. O., Martínez-Carranza, J., & Cruz-Vega, I. (2018) · International Conference on Electronics, Communications, and Computers · doi:10.1109/conielecomp.2018.8327189 · Cited by 49
Confronting machine-learning with neuroscience for neuromorphic architectures design
Khacef, L., Abderrahmane, N., & Miramond, B. (2018) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2018.8489241 · Cited by 51
“the input neurons only transmit the information to the first hidden layer), one or more hidden layers (called Deep Neural Network, or DNN, for a large number of hidden layers), and an output layer [2].”
BackgroundMethodConti, F., Cavigelli, L., Paulin, G., Susmelj, I., & Benini, L. (2018) · Custom Integrated Circuits Conference · doi:10.1109/cicc.2018.8357068 · Cited by 46
Mind your privacy: Privacy leakage through BCI applications using machine learning methods
Landau, O., Cohen, A., Gordon, S. C., & Nissim, N. (2020) · Knowledge-Based Systems · doi:10.1016/j.knosys.2020.105932 · Cited by 51
ESP4ML: Platform-Based Design of Systems-on-Chip for Embedded Machine Learning
Giri, D., Chiu, K.-L., Guglielmo, G. D., Mantovani, P., & Carloni, L. P. (2020) · Design, Automation, and Test in Europe · doi:10.23919/date48585.2020.9116317 · Cited by 60
“Since 2012, when the use of deep neural networks for classifying million of images from the web gave spectacular results [1], [2], the design of specialized accelerators for machine learning (ML) has become the main trend across all types of computing systems [3].”
BackgroundBoosting Bit-Error Resilience of DNN Accelerators Through Median Feature Selection
Ozen, E., & Orailoğlu, A. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.3012209 · Cited by 46
“For instance, the frequent utilization of DNN optimization techniques [7] (e.g., channel pruning [50]) necessitates the consideration of their interaction with the median feature selection.”
BackgroundMethodInfluentialYield performance estimation of corn hybrids using machine learning algorithms
Sarijaloo, F. B., Porta, M., Taslimi, B., & Pardalos, P. M. (2021) · Artificial Intelligence in Agriculture · doi:10.1016/j.aiia.2021.05.001 · Cited by 60
Elastic-DF: Scaling Performance of DNN Inference in FPGA Clouds through Automatic Partitioning
Alonso, T., Petrică, L., Ruiz, M., Petri-Koenig, J., Umuroglu, Y., Stamelos, I., Koromilas, E., Blott, M., & Vissers, K. (2021) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3470567 · Cited by 50
Cong, Y., & Endo, T. (2022) · OMICS A Journal of Integrative Biology · doi:10.1089/omi.2022.0068 · Cited by 52
“Deep neural networks (DNNs) increase the number of layers based on NN, optimizes the model structure and training methods, and improves the computational power to better cope with large amounts of data (Sze et al, 2017).”
Data Shuffling in Wireless Distributed Computing via Low-Rank Optimization
Yang, K., Shi, Y., & Ding, Z. (2019) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2019.2912139 · Cited by 43
American Sign Language Posture Understanding with Deep Neural Networks
Jalal, M. A., Chen, R., Moore, R. K., & Mihaylova, L. S. (2018) · International Conference on Information Fusion (FUSION) · doi:10.23919/icif.2018.8455725 · Cited by 44
“Convolutional Neural Networks (CNNs) have been very useful for understanding gestures and events [18].”
BackgroundStructured Pruning for Efficient ConvNets via Incremental Regularization
Wang, H., Zhang, Q., Wang, Y., Yu, L., & Hu, H. R. (2019) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2019.8852463 · Cited by 58
“To avoid 17 hardware-unfriendly irregular sparsity, structured pruning is proposed for CNN acceleration [1, 23].”
MethodMousouliotis, P., & Petrou, L. (2020) · Microprocessors and Microsystems · doi:10.1016/j.micpro.2020.102990 · Cited by 54
Energy-Efficient Machine Learning on the Edges
Kumar, M., Zhang, X., Liu, L., Wang, Y., & Shi, W. (2020) · IEEE International Symposium on Parallel & Distributed Processing, Workshops and Phd Forum · doi:10.1109/ipdpsw50202.2020.00153 · Cited by 62
“Generally, the co-design approaches can be divided into two categories [44]:”
BackgroundOMNI: A Framework for Integrating Hardware and Software Optimizations for Sparse CNNs
Liang, Y. E., Lu, L., & Xie, J. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.3023903 · Cited by 47
“Due to their intensive computation, convolutional layers are always the computation bottleneck in CNNs[41].”
BackgroundEnergy-efficient deep learning inference on edge devices
Daghero, F., Pagliari, D. J., & Poncino, M. (2020) · Advances in computers · doi:10.1016/bs.adcom.2020.07.002 · Cited by 49
EdgeKE: An On-Demand Deep Learning IoT System for Cognitive Big Data on Industrial Edge Devices
Fang, W. W., Feng, X., Ding, Y., Xiong, N. N., & Leung, V. C. M. (2020) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2020.3044930 · Cited by 48
“However, these unstructured pruning methods result in sparse weight matrices, which cannot function without dedicated hardware and libraries [4].”
BackgroundDayal, A., Paluru, N., Cenkeramaddi, L. R., Soumya, J., & Yalavarthy, P. K. (2021) · Electronics · doi:10.3390/electronics10020182 · Cited by 46
“These networks are being applied in a variety of applications, such as object detection tasks, speech recognition tasks, unmanned aerial vehicles (UAVs), and unmanned ground vehicles (UGVs) [15].”
BackgroundIn situ Parallel Training of Analog Neural Network Using Electrochemical Random-Access Memory
Li, Y., Xiao, T. P., Bennett, C. H., Isele, E., Melianas, A., Tao, H., Marinella, M. J., Salleo, A., Fuller, E. J., & Talin, A. A. (2021) · Frontiers in Neuroscience · doi:10.3389/fnins.2021.636127 · Cited by 53
“The energy intensity arises from the need to shuttle large quantities of information from memory to processor to conduct large matrix multiplications and to update matrix weights (Marinella et al., 2017; Sze et al., 2017).”
BackgroundA Comprehensive Survey of Machine Learning Applied to Radar Signal ProcessingPreprint
Lang, P., Fu, X., Martorella, M., Dong, J., Qin, R., Meng, X., & Xie, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2009.13702 · Cited by 80
“Therefore, the design of lightweight models or efficient DNN models [39] is necessary for DL model to be efficiently applied to RSP domain.”
BackgroundMethodIR-QNN Framework: An IR Drop-Aware Offline Training of Quantized Crossbar Arrays
Fouda, M. E., Lee, S., Lee, J., Kim, G. H., Kurdahi, F. J., & Eltawil, A. M. (2020) · IEEE Access · doi:10.1109/access.2020.3044652 · Cited by 42
“INTRODUCTION Artificial Intelligence hardware acceleration has attracted significant interest [1], [2] especially accelerating deep neural networks (DNNs) with in-memory processing, alleviating the memory-wall bottleneck problem in the Von-Neumann computing architecture.”
BackgroundTowards explainable AI for hyperspectral image classification in Edge Computing environments
Lucia, G. D., Lapegna, M., & Romano, D. (2022) · Computers & Electrical Engineering · doi:10.1016/j.compeleceng.2022.108381 · Cited by 49
Achu, A. L., Thomas, J., Aju, C. D., Remani, P. K., & Gopinath, G. (2022) · Earth Science Informatics · doi:10.1007/s12145-022-00910-8 · Cited by 45
Transfer Learning in Sensor-Based Human Activity Recognition: A Survey
Dhekane, S. G., & Ploetz, T. (2025) · ACM Computing Surveys · doi:10.1145/3717608 · Cited by 42
“Considering computations, most multiply-accumulate (MAC) operations take place in the fully connected and convolutional layers [198], resulting in a large requirement of power/battery.”
Luque, J., Cortès, G., Segura, C., Maravilla, A., Esteban, J., & Fabregat, J. (2018) · European Signal Processing Conference (EUSIPCO) · doi:10.23919/eusipco.2018.8553585 · Cited by 42
“trial lead to a number of MAC operations [20] around 26K, that translates into roughly 20ms for a Raspberry in order to perform person authentication every second.”
BackgroundMnnFast: a fast and scalable system architecture for memory-augmented neural networks
Jang, H., Kim, J., Jo, J.-E., Lee, J., & Kim, J. (2019) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1145/3307650.3322214 · Cited by 84
“Spatial architecture is optimized for DNNs, as shared values are broadcasted toALUs, and intermediate resultsflow betweenALUs [70].”
BackgroundFPGA Logic Block Architectures for Efficient Deep Learning Inference
Eldafrawy, M. M., Boutros, A., Yazdanshenas, S., & Betz, V. (2020) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3393668 · Cited by 49
Electrical equipment identification in infrared images based on ROI-selected CNN method
Han, S., Yang, F., Yang, G., Gao, B., Zhang, N., & Wang, D. (2020) · Electric Power Systems Research · doi:10.1016/j.epsr.2020.106534 · Cited by 42
Marchisio, A., Massa, A., Mrazek, V., Bussolino, B., Martina, M., & Shafique, M. (2020) · 2020 IEEE/ACM International Conference On Computer Aided Design (ICCAD) · doi:10.1145/3400302.3415731 · Cited by 45
“Highly-accurate DNNs require a significant amount of hardware resources, which may not be feasible on resource-constrained IoTEdge devices [2][18][28][32].”
BackgroundConvolution Inference via Synchronization of a Coupled CMOS Oscillator Array
Nikonov, D. E., Kurahashi, P., Ayers, J. S., Li, H., Kamgaing, T., Dogiamis, G. C., Lee, H., Fan, Y., & Young, I. A. (2020) · IEEE Journal on Exploratory Solid-State Computational Devices and Circuits · doi:10.1109/jxcdc.2020.3046143 · Cited by 42
“The most compute-intensive operation in CNN is a convolution of an image with a set of kernels.”
MethodInfluentialIntermittent-Aware Neural Architecture Search
Mendis, H. R., Kang, C.-K., & Hsiu, P.-C. (2021) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3476995 · Cited by 52
“Furthermore, advanced models with skip connections and feedback paths [58] can also be supported by including the additional data fetching costs in the progress recovery cost.”
BackgroundMethodInfluentialWei, L., Zhang, Y., Lu, Q., Yuan, Z., Li, H., & Huang, Q. (2021) · Ecological Indicators · doi:10.1016/j.ecolind.2021.108384 · Cited by 43
Samanta, R. K., Sadhukhan, B., Samaddar, H., Sarkar, S., Koner, C., & Ghosh, M. (2022) · CAAI Transactions on Intelligence Technology · doi:10.1049/cit2.12114 · Cited by 53
Double-kernel based class-specific broad learning system for multiclass imbalance learning
Chen, W., Yang, K., Yu, Z., & Zhang, W. (2022) · Knowledge-Based Systems · doi:10.1016/j.knosys.2022.109535 · Cited by 43
Accelerating Neural Network Inference With Processing-in-DRAM: From the Edge to the Cloud
Junior, G. F. D. O., Gómez-Luna, J., Ghose, S., Boroumand, A., & Mutlu, O. (2022) · IEEE Micro · doi:10.1109/mm.2022.3202350 · Cited by 64
Qi, C., Wu, M., Liu, H., Liang, Y., Liu, X., & Lin, Z. (2023) · Journal of Cleaner Production · doi:10.1016/j.jclepro.2023.136771 · Cited by 44
Lee, S., Lee, Y., Kim, Y., Han, Y., Kwon, H.-H., & Hong, D. (2023) · Journal of Guidance Control and Dynamics · doi:10.2514/1.g007715 · Cited by 41
Energy Consumption of Machine Learning Enhanced Open RAN: A Comprehensive Review
Liang, X., Wang, Q., Al-Tahmeesschi, A., Chetty, S. B., Grace, D., & Ahmadi, H. (2024) · IEEE Access · doi:10.1109/access.2024.3412758 · Cited by 62
“DNNs are sometimes called Multiple-layer Perceptron (MLP) [48].”
BackgroundSakemi, Y., Morino, K., Morie, T., & Aihara, K. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3095068 · Cited by 42
Automatic Design of Convolutional Neural Network Architectures Under Resource Constraints
Li, S., Sun, Y., Yen, G. G., & Zhang, M. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3123105 · Cited by 54
“There is a common belief that deeper CNN models typically perform better than shallower ones [4], [5].”
BackgroundPruning Algorithms to Accelerate Convolutional Neural Networks for Edge Applications: A SurveyPreprint
Liu, J., Tripathi, S., Kurup, U., & Shah, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2005.04275 · Cited by 78
“Quantization [41] is another popular compression technique which involves replacing high precision floating points in CNN weight matrices with lower precision representations.”
MethodRandom sketch learning for deep neural networks in edge computing
Li, B., Chen, P., Liu, H., Guo, W., Cao, X., Du, J., Zhao, C., & Zhang, J. (2021) · Nature Computational Science · doi:10.1038/s43588-021-00039-6 · Cited by 40
Locality-Centric Data and Threadblock Management for Massive GPUs
Khairy, M., Nikiforov, V., Nellans, D. W., & Rogers, T. G. (2020) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1109/micro50266.2020.00086 · Cited by 42
“We used the optimized sgemm from [57], [75] as our reference implementation of GEMM and we extract layer and matrices dimensions from several popular DL networks [4], [25], [77].”
MethodJunjia, Y., Alias, A. H., Haron, N. A., & Bakar, N. A. (2023) · Sustainability · doi:10.3390/su151511803 · Cited by 61
Review of neural network model acceleration techniques based on FPGA platforms
Liu, F., Li, H., Hu, W., & He, Y. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.128511 · Cited by 53
Joint Signal Detection and Automatic Modulation Classification via Deep Learning
Xing, H., Zhang, X., Chang, S., Ren, J., Zhang, Z., Xu, J., & Cui, S. R. (2024) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2024.3450972 · Cited by 40
“In recent years, deep learning (DL) has achieved astonishing success in object detection and classification [10].”
BackgroundCNN-Based LOS/NLOS Identification in 3-D Massive MIMO Systems
Zeng, T., Chang, Y., Zhang, Q., Hu, M., & Li, J. (2018) · IEEE Communications Letters · doi:10.1109/lcomm.2018.2872522 · Cited by 41
“Although from [1] we can infer that LOS/NLOS conditions have complex non-linear relationship with these information, convolutional neural network (CNN) can help, which has achieved spectacular success in feature extractors, inverse problems and image classification [7], [8].”
Analysis of Efficient CNN Design Techniques for Semantic Segmentation
Briot, A., Viswanath, P., & Yogamani, S. K. (2018) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw.2018.00109 · Cited by 48
“An overview of efficient CNN from a hardware perspective is provided in [36] and guidelines for design of small networks is provided in [16].”
MethodFerroelectric FET Based In-Memory Computing for Few-Shot Learning
Laguna, A. F., Yin, X., Reis, D. A., Niemier, M. T., & Hu, X. S. (2019) · ACM Great Lakes Symposium on VLSI (GLSVLSI) · doi:10.1145/3299874.3319450 · Cited by 40
Srinivas, K., & Dua, M. (2019) · Lecture notes in networks and systems · doi:10.1007/978-3-030-33846-6_69 · Cited by 40
RISC-V2: A Scalable RISC-V Vector Processor
Patsidis, K., Nicopoulos, C. A., Sirakoulis, G. C., & Dimitrakopoulos, G. (2020) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas45731.2020.9181071 · Cited by 64
“Consequently, they are not tuned to effectively handle the massively data-parallel workloads that machine learning algorithms and NNs have brought to the forefront [3].”
BackgroundWu, J., Mo, F., Saraya, T., Hiramoto, T., & Kobayashi, M. (2020) · IEEE Transactions on Electron Devices · doi:10.1109/ted.2020.3033831 · Cited by 42
“IN-MEMORY computing shows a large advantage of high energy efficiency for deep neural network application [1].”
BackgroundA Divide-and-Conquer Genetic Programming Algorithm With Ensembles for Image Classification
Bi, Y., Xue, B., & Zhang, M. (2021) · IEEE Transactions on Evolutionary Computation · doi:10.1109/tevc.2021.3082112 · Cited by 43
Enabling Training of Neural Networks on Noisy Hardware
Gokmen, T. (2021) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2021.699148 · Cited by 51
Selvaggio, A. Z., Sousa, F. M. M., da Silva, F. V., & Vianna, S. S. V. (2022) · Process Safety and Environmental Protection · doi:10.1016/j.psep.2022.01.021 · Cited by 40
Jing, Z., Yan, B., Yang, Y., & Huang, R. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3186024 · Cited by 40
Lee, J., Choi, J., Kim, J., Lee, J.-H., & Kim, Y. (2021) · Design Automation Conference · doi:10.1109/dac18074.2021.9586312 · Cited by 43
“Among the layers, convolutional layers have been a primary acceleration target of NPUs as they tend to incur the largest amounts of computation [15].”
BackgroundMethodInfluentialRadar Signal Recognition Based on CNN With a Hybrid Attention Mechanism and Skip Feature Aggregation
Guo, Y., Sun, H., Liu, H., & Deng, Z. (2022) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2022.3204100 · Cited by 39
“Among many natural image processing methods, such as image classification, image style transfer, image super-resolution, and text image conversion [12], [13], [14], it shows extraordinary potential.”
BackgroundPower Allocation and User Grouping for NOMA Downlink Systems
Li, J., Gao, T., He, B., Zheng, W., & Lin, F. (2023) · Applied Sciences · doi:10.3390/app13042452 · Cited by 44
“The deep neural network (DNN) is one of the most basic and widely used networks in DL [25].”
Efficient Privacy-Preserving Inference Outsourcing for Convolutional Neural Networks
Yang, X., Chen, J., He, K., Bai, H., Wu, C., & Du, R. (2023) · IEEE Transactions on Information Forensics and Security · doi:10.1109/tifs.2023.3287072 · Cited by 42
“According to the calculation involved, there are the following four types of layers [29].”
BackgroundExplainable AI models for predicting drop coalescence in microfluidics device
Hu, J., Zhu, K., Cheng, S. L., Kovalchuk, N. M., Soulsby, A., Simmons, M. J. H., Matar, O. K., & Arcucci, R. (2024) · Chemical Engineering Journal · doi:10.1016/j.cej.2023.148465 · Cited by 48
“…their capacity to adeptly handle various types of data has fostered their wide application across diverse tasks, such as machine translation, sentiment analysis, image and speech recognition, anomaly detection, text classification, as well as various regression and classification problems [70,71].”
NN-Noxim: High-Level Cycle-Accurate NoC-based Neural Networks Simulator
Chen, K.-C. J., & Wang, T. (2018) · International Workshop on Network on Chip Architectures (NoCArc) · doi:10.1109/nocarc.2018.8541173 · Cited by 39
“In addition, the GAN Playground is unable to provide the hardware analysis either.”
MethodInfluentialAutomated design of error-resilient and hardware-efficient deep neural networks
Schorn, C., Elsken, T., Vogel, S., Runge, A., Guntoro, A., & Ascheid, G. (2020) · Neural Computing and Applications · doi:10.1007/s00521-020-04969-6 · Cited by 48
NEWLSTM: An Optimized Long Short-Term Memory Language Model for Sequence Prediction
Wang, Q., Peng, R.-Q., Wang, J., Li, Z., & Qu, H. (2020) · IEEE Access · doi:10.1109/access.2020.2985418 · Cited by 43
“Its hidden layer update is replaced by a dedicated memory unitCt , which performs better in the face of longer sequences and is suitable for context-sensitive language learning [27].”
BackgroundWu, N., Jiang, T., Zhang, L., Zhou, F., & Ge, F. (2020) · Electronics · doi:10.3390/electronics9061005 · Cited by 49
“In Reference [7], the matrix multiplication in the CNN algorithm is reduced by the FFT, but the FFT itself consumes more hardware resources.”
MethodVerifiable and privacy preserving federated learning without fully trusted centers
Han, G., Zhang, T., Zhang, Y., Xu, G., Sun, J., & Cao, J. P. (2021) · Journal of Ambient Intelligence and Humanized Computing · doi:10.1007/s12652-020-02664-x · Cited by 51
Koziel, S. M., Mahoutı, P., Çalık, N., Belen, M. A., & Szczepański, S. (2021) · IEEE Access · doi:10.1109/access.2021.3078432 · Cited by 43
DSP-Packing: Squeezing Low-precision Arithmetic into FPGA DSP Blocks
Sommer, J., Özkan, M. A., Keszöcze, O., & Teich, J. (2022) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl57034.2022.00035 · Cited by 39
“In contrast, standard Neural Networks (NNs) are based on Multiply-Accumulate operations (MAC) [14].”
BackgroundValue-laden disciplinary shifts in machine learning
Dotan, R., & Milli, S. (2020) · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · doi:10.1145/3351095.3373157 · Cited by 78
Song, Y., Wang, X., Wu, Q., Yang, F., Wang, C., Wang, M., & Miao, X. (2022) · Advanced Science · doi:10.1002/advs.202200036 · Cited by 37
To Spike or Not to Spike: A Digital Hardware Perspective on Deep Learning Acceleration
Ottati, F., Gao, C., Chen, Q., Brignone, G., Casu, M. R., Eshraghian, J. K., & Lavagno, L. (2023) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2023.3330432 · Cited by 56
“In ANN architectures, the most common operation is the multiply and accumulate (MAC) [19], which includes the input activation multiplication by the corresponding weight and the subsequent accumulation.”
BackgroundXie, S., Ni, C., Sayal, A., Jain, P., Hamzaoglu, F., & Kulkarni, J. P. (2023) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2023.3326094 · Cited by 38
“For example, a basic network, like LeNET-5 requires a 341 k number of MAC operations, while a more sophisticated network requires a 15.5 G number of MAC operations [6].”
BackgroundMethodInfluentialChen, C.-W. (2024) · Sustainability · doi:10.3390/su16114432 · Cited by 59
A Systematic Review for Classification and Selection of Deep Learning Methods
Saputra, N. A., Riza, L. S., Setiawan, A., & Hamidah, I. (2024) · Decision Analytics Journal · doi:10.1016/j.dajour.2024.100489 · Cited by 57
Deep Learning in Power Systems: A Bibliometric Analysis and Future Trends
Miraftabzadeh, S. M., Martino, A. D., Longo, M., & Zaninelli, D. (2024) · IEEE Access · doi:10.1109/access.2024.3491914 · Cited by 50
“Emerging methods such as PINNs, BNNs and GNNs are being implemented in various applications, including language modeling, reading comprehension, and question answering through supervised learning and fine-tuning [7], [8].”
BackgroundMethodSub-Word Parallel Precision-Scalable MAC Engines for Efficient Embedded DNN Inference
Mei, L., Dandekar, M., Rodopoulos, D., Constantin, J., Debacker, P., Lauwereins, R., & Verhelst, M. (2019) · IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS) · doi:10.1109/aicas.2019.8771481 · Cited by 47
“Yet, it has been shown that there is a wide variety of precision requirements among (non-binary) neural networks [10], [11], ranging from 2 to 16 bit.”
BackgroundEvaluation of maxout activations in deep learning across several big data domains
Castaneda, G., Morris, P., & Khoshgoftaar, T. M. (2019) · Journal Of Big Data · doi:10.1186/s40537-019-0233-0 · Cited by 41
Maddah, M., Mandegar, M. A., Dame, K., Grafton, F., Loewke, K. E., & Ribeiro, A. J. S. (2020) · Journal of Pharmacological and Toxicological Methods · doi:10.1016/j.vascn.2020.106895 · Cited by 37
Wang, S., Chen, X., Huang, X., Zhang, D. W., & Zhou, P. (2020) · Advanced Intelligent Systems · doi:10.1002/aisy.202000124 · Cited by 44
Non‐invasive setup for grape maturation classification using deep learning
Ramos, R. P., Gomes, J. S., Prates, R. M., Filho, E. F. D. S., Teruel, B., & Costa, D. D. S. (2020) · Journal of the Science of Food and Agriculture · doi:10.1002/jsfa.10824 · Cited by 38
Modeling and predicting vehicle accident occurrence in Chattanooga, Tennessee
Roland, J., Way, P., Firat, C. B., Doan, T.-N., & Sartipi, M. (2020) · Accident Analysis & Prevention · doi:10.1016/j.aap.2020.105860 · Cited by 50
DynO: Dynamic Onloading of Deep Neural Networks from Cloud to Device
Almeida, M., Laskaridis, S., Venieris, S. I., Leontiadis, I., & Lane, N. D. (2022) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3510831 · Cited by 52
“In terms of how to quantize, existing works typically adopt linear quantization based on block loating-point (also known as dynamic ixed-point) [14, 15, 25, 33, 56] which dictates a uniform bitwidth and a non-uniform scale factor across the model’s layers.”
BackgroundDual-Tree Genetic Programming for Few-Shot Image Classification
Bi, Y., Xue, B., & Zhang, M. (2021) · IEEE Transactions on Evolutionary Computation · doi:10.1109/tevc.2021.3100576 · Cited by 40
“In recent years, CNNs, which can automatically learn informative features from images, have been widely applied to image classification and achieved promising results [3], [27].”
BackgroundRon, D. A., Freire, P. J., Prilepsky, J. E., Kamalian-Kopae, M., Napoli, A., & Turitsyn, S. K. (2022) · Scientific Reports · doi:10.1038/s41598-022-12563-0 · Cited by 40
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.