Model Compression and Hardware Acceleration for Neural Networks: A Comprehensive Survey
The DOI matches its Crossref record; 317 of the 334 references it cites were found in a registry or an index, and 1 are not checked yet.
Abstract
Domain-specific hardware is becoming a promising topic in the backdrop of improvement slow down for general-purpose processors due to the foreseeable end of Moore's Law. Machine learning, especially deep neural networks (DNNs), has become the most dazzling domain witnessing successful applications in a wide spectrum of artificial intelligence (AI) tasks. The incomparable accuracy of DNNs is achieved by paying the cost of hungry memory consumption and high computational complexity, which greatly impedes their deployment in embedded systems. Therefore, the DNN compression concept was naturally proposed and widely used for memory saving and compute acceleration. In the past few years, a tremendous number of compression techniques have sprung up to pursue a satisfactory tradeoff between processing efficiency and application accuracy. Recently, this wave has spread to the design of neural network accelerators for gaining extremely high performance. However, the amount of related works is incredibly huge and the reported approaches are quite divergent. This research chaos motivates us to provide a comprehensive survey on the recent advances toward the goal of efficient compression and execution of DNNs without significantly compromising accuracy, involving both the high-level algorithms and their applications in hardware design. In this article, we review the mainstream compression approaches such as compact model, tensor decomposition, data quantization, and network sparsification…
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
“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 [28]
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 3×
“Specifically, an example of the identity connection used in ResNet [30] [see Fig.”
BackgroundMethodVery 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 1×
“In VGG-Nets [27], the interlayer topology is rather simple, where multiple Conv layers are stacked head-to-tail in sequence.”
BackgroundImageNet 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 3×
“1078 [20] A.”
BackgroundInfluentialHochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318 · cited 2×
“…we discuss recent advances in RNNs that aim to improve the execution efficiency by designing compact RNNs in the level of unit or network. a) Unit level: LSTM is designed to improve the long-term dependence over vanilla RNNs by alleviating the issue of gradient explosion or vanishing [18], [46].”
BackgroundMethodLeCun, Y., Bengio, Y., & Hinton, G. (2015) · Nature · doi:10.1038/nature14539 · Cited by 84,126
Densely Connected Convolutional Networks
Huang, G., Liu, Z., Van Der Maaten, L., & Weinberger, K. Q. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.243 · Cited by 48,566
Going 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
(2014) · Cited by 33,419
Rethinking the Inception Architecture for Computer Vision
Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., & Wojna, Z. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.308 · Cited by 32,138
MobileNetV2: Inverted Residuals and Linear Bottlenecks
Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., & Chen, L.-C. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00474 · Cited by 28,755 · cited 35×
“The feature propagation gradually improves the abstraction level from local to global, which finally enables the network decision such as image recognition. a) Spatial correlation to enlarge RF: Compared to MLP, the key feature of CNNs is the local connections with shared weights, which results in high spatial correlat…”
BackgroundMethodInfluential
Research works printed but not linked: 2
- lin, 2013, Network in network, arXiv 1312 4400
- han, 2015, Learning both weights and connections for efficient neural network, Proc Adv Neural Inf Process Syst, 1135
Cited by
Items 701–800 of 1,002
Distributed Clock Phase and Frequency Synchronization in Half-Duplex TDMA Networks
Zino, I., Dabora, R., & Poor, H. V. (2025) · IEEE Transactions on Communications · doi:10.1109/tcomm.2025.3606416 · Cited by 2
“We note that, since each DNN is used for inference only once per 3 · N time slots, and there are two DNNs at each node, the rate of computation at a node is 1500 products for inference per TDMA frame, which is a feasible computational load for real-time modern microcontrollers, see [11] and [28].”
Optimizing and Exploring System Performance in Compact Processing-in-Memory-based Chips
Chen, P., & Yang, X. (2025) · IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS) · doi:10.1109/aicas64808.2025.11173147 · Cited by 2
“Convolutional neural networks (CNNs) have gained great success in various applications such as image recognition and autonomous driving [1]–[3].”
Model quantization for computing-in-memory: a survey
Sun, S., Bai, J., Chen, H., Deng, K., Xie, Z., Li, J., Cao, B., Zhang, H., Kang, W., & Zhao, W. S. (2025) · Science China Information Sciences · doi:10.1007/s11432-024-4522-8 · Cited by 3
“Meanwhile, hardware-related reviews mainly explore the implementation of QNNs and hardware-aware optimizations across various platforms, including central processing units (CPUs), GPUs, FPGAs, and traditional application-specific integrated circuits (ASICs) [32–35].”
Gao, C., Chen, X., Wang, J., Wang, J., & Yang, G. (2025) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2025.111914 · Cited by 3
Weight-Rounding Error in Deep Neural Networks
Šíma, J., & Vidnerová, P. (2025) · Lecture notes in computer science · doi:10.1007/978-3-032-06078-5_23 · Cited by 2
Topographical sparse mapping: A neuro-inspired sparse training framework for deep learning models
Rad, M. K., Neri, F., Moschoyiannis, S., & Bauer, R. (2025) · Neurocomputing · doi:10.1016/j.neucom.2025.131740 · Cited by 2
MOSDA: On-Chip Memory Optimized Sparse Deep Neural Network Accelerator With Efficient Index Matching
Xu, H., Shiomi, J., & Onodera, H. (2020) · IEEE Open Journal of Circuits and Systems · doi:10.1109/ojcas.2020.3035402 · Cited by 2
“As DNN networks become deeper and more complex, the required computing power and energy consumption are also increasing [7]–[9].”
BackgroundIdleSR: Efficient Super-Resolution Network with Multi-scale IdleBlocks
Xiong, D., Huang, K., Jiang, H., Li, B., Chen, S., & Jiang, X. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-67070-2_8 · Cited by 2
Web Based GPU Acceleration in Embodied Agent Training Workflow
Николаевич, С. А., Dobrynin, V., & Yulmetova, O. S. (2021) · IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering · doi:10.1109/elconrus51938.2021.9396663 · Cited by 2
Tampubolon, H., Setyoko, A. S., & Purnamasari, F. (2021) · Journal of Physics Conference Series · doi:10.1088/1742-6596/1898/1/012038 · Cited by 2
Choi, Y., El‐Khamy, M., & Lee, J. (2021) · Proceedings - International Conference on Image Processing · doi:10.1109/icip42928.2021.9506776 · Cited by 2
“Network quantization is an important procedure for efficient inference when deploying pre-trained deep neural networks on resource-limited platforms [4].”
MethodModel compression for on-device inference
Shi, Y., Yang, K., Yang, Z., & Zhou, Y. (2021) · Elsevier eBooks · doi:10.1016/b978-0-12-823817-2.00015-2 · Cited by 2
Li, D., Ke, Z., & Zhou, X. (2021) · ACM Symposium on QoS and Security for Wireless and Mobile Networks (Q2SWinet) · doi:10.1145/3479242.3487322 · Cited by 3
“(i) A natural approach is to run the object detection tasks on mobile devices by using model compression techniques, such as weight sharing [1], model pruning [2], and knowledge distillation [2] to turn the workload-heavy CNN models into light-weight ones.”
An Adaptive Logarithm Quantization Method for DNN Compression
Wang, Y., He, Z., Tang, C., Wang, Z., & Zhu, W. (2021) · Communications in computer and information science · doi:10.1007/978-3-030-92307-5_41 · Cited by 3
Chebyshev Polynomial Broad Learning System
Feng, S., Wang, B., & Chen, C. L. P. (2021) · International Conference on Information, Cybernetics, and Computational Social Systems (ICCSS) · doi:10.1109/iccss53909.2021.9721962 · Cited by 2
A data-aware dictionary-learning based technique for the acceleration of deep convolutional networks
Pikoulis, E., Mavrokefalidis, C., & Lalos, A. S. (2021) · IEEE International Workshop on Multimedia Signal Processing (MMSP) · doi:10.1109/mmsp53017.2021.9733527 · Cited by 2
“As high-end DNN models consist of tens, or even hundreds of millions of parameters [2], the design of smaller/faster, still high-performing, DNN models has been recently considered a key objective.”
BackgroundFACCU: Enable Fast Accumulation for High-Speed DSP Systems
Wang, M., Cheng, X., Zou, D., & Wang, Z. (2022) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2022.3196398 · Cited by 4
Bao, H., Qin, Y., Chen, J., Yang, L., Li, J., Zhou, H., Li, Y., & Miao, X. (2022) · Frontiers in Electronics · doi:10.3389/felec.2022.954661 · Cited by 3
“The quantization, which has been widely used in industry and academia to compress the neural networks (Sze et al., 2017; Deng et al., 2020), is not only intrinsically embedded within the”
BackgroundGarg, S., & Singh, P. (2022) · IEEE Bombay Section Signature Conference (IBSSC) · doi:10.1109/ibssc56953.2022.10037360 · Cited by 2
“Model compression and acceleration techniques have already attracted researchers inthe last couple of years and many techniques are gradually developed in this area, some of them are, low-rank factorization, knowledge distillation, network pruning, and quantization[9][10].”
BackgroundMulti-Head Convolutional Neural Network Compression based on High-Order Principal Component Analysis
Kim, T., Na, Y., & Park, S. (2023) · International Conference on Electronics, Information, and Communication (ICEIC) · doi:10.1109/iceic57457.2023.10049909 · Cited by 2
“However, these peculiarities also create implementation challenges when deploying multi-head convolutional neural network on resource-constrained devices, such as mobile phones and embedded systems [3].”
BackgroundFast data-free model compression via dictionary-pair reconstruction
Gao, Y., Zhang, Z., Zhang, H., Zhao, M., Yang, Y., & Wang, M. (2023) · Knowledge and Information Systems · doi:10.1007/s10115-023-01846-1 · Cited by 2
“Pruning-based methods produce sparsity in DNNs [34] to reduce the calculation and storage by removing the unimportant components at different levels, e.g., filters [18], connections [35], channels [5] and layers [36].”
Method3D-KCPNet: Efficient 3DCNNs based on tensor mapping theory
Lv, R., Wang, D., Zheng, J., & Yang, Z.-X. (2023) · Neurocomputing · doi:10.1016/j.neucom.2023.126985 · Cited by 2
Pimentel, A., Guimarães, H. R., Avila, A. R., Rezagholizadeh, M., & Falk, T. H. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2309.14462 · Cited by 2
“The method can be further extended to represent gradient and activation in the quantized form [7].”
MethodLi, J., Magar, R. T., Chen, D., Lin, F., Wang, D., Yin, X., Zhuang, W., & Li, Z. (2023) · SSRN Electronic Journal · doi:10.2139/ssrn.4639580 · Cited by 2
A New Lightweight Network for Real-time Detection of Railway InfrastructurePreprint
Cheng, Y., & Cheng, R. (2024) · Research Square · doi:10.21203/rs.3.rs-3932509/v1 · Cited by 2
Deng, L., Tang, H., & Roy, K. (2024) · Frontiers in Computational Neuroscience · doi:10.3389/fncom.2024.1455530 · Cited by 2
“Spiking neural networks (SNNs) and neuromorphic chips (Basu et al., 2022; Christensen et al., 2022) dominate the neuromorphic computing domain, while artificial neural networks (ANNs) and machine learning accelerators (Deng et al., 2020) dominate the machine learning domain.”
BackgroundDynaSplit: A Hardware-Software Co-Design Framework for Energy-Aware Inference on EdgePreprint
May, D., Tundo, A., Ilager, S., & Brandić, I. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2410.23881 · Cited by 3
Meta-Sparsity: Learning Optimal Sparse Structures in Multi-task Networks through Meta-learningPreprint
Upadhyay, R., Phlypo, R., Saini, R., & Liwicki, M. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2501.12115 · Cited by 2
“It facilitates the practical application of DL models in real-world scenarios, fostering sustainability and enhancing accessibility on a broader scale (Deng et al. (2020)).”
A Hardware-Efficient Spiking Predictive Coding Architecture with Online STDP LearningPreprint
Kim, J.-G., Seo, D., Yeo, I., & Lee, B. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5699160 · Cited by 2
Hardware Aspects of Parallel Neural Network Implementation
Kouretas, I., & Paliouras, V. (2021) · International Conference on Modern Circuits and Systems Technologies · doi:10.1109/mocast52088.2021.9493365 · Cited by 1
A Survey on System-Level Design of Neural Network Accelerators
Seto, K. (2021) · Journal of Integrated Circuits and Systems · doi:10.29292/jics.v16i2.505 · Cited by 2
“Network Architecture Search (NAS) tries to find accurate, compact and hardware friendly CNN model automatically [1][6].”
BackgroundMethodInfluentialVariance-Guided Structured Sparsity in Deep Neural Networks
Pandit, M. K., & Banday, M. (2022) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2022.3221688 · Cited by 1
“ADMM [63] achieves 30 × sparsity with 56.8% (top-1) and 80.2% (top-5) accuracy.”
InfluentialBailey, D. G. (2023) · Wiley eBooks · doi:10.1002/9781119819820.ch14 · Cited by 1
Adaptive Splitting Algorithm for Neural Network Modelling in AMP
Zhang, K., Ma, Y., Wu, Y., Chen, T., & Ma, H. (2023) · IEEE International Symposium on Parallel and Distributed Processing with Applications (ISPA) · doi:10.1109/ispa-bdcloud-socialcom-sustaincom59178.2023.00059 · Cited by 1
“Deng et al. argued that DNN compression can be widely used to save memory and accelerate computation [5].”
BackgroundFan, F., Zhang, M., Liu, Y., Yu, D., & Li, J. (2024) · Measurement Science and Technology · doi:10.1088/1361-6501/ad82ff · Cited by 1
“In contrast, parameter compression methods aim to reduce the storage requirements and computational overhead of a model through techniques such as pruning, quantization, and knowledge distillation (KD) [11].”
Sub-model Parallelism: A Scale-out Deployment Method for Large Multi-modal DNNs
Huang, T., Sun, L., Hou, X., Zhu, X., Xia, X., Wang, Y., Chen, M., & Li, C. (2024) · IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid) · doi:10.1109/ccgrid59990.2024.00081 · Cited by 1
“For a single device, model trimming technologies including quantization, compression, pruning and knowledge distillation have experienced fast development [22], [24].”
BackgroundLow-Complexity Neural Belief Propagation Decoding Algorithm Based on Tensor Ring Decomposition
Liang, Y., Lam, C., Wu, Q., Ng, B. K., & Im, S. (2024) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2024.3487999 · Cited by 1
“Compared with other compression methods in intelligent communication, the compression method based on tensor decomposition has better interpretability and rigorous mathematical principles [16], [20].”
Evaluation and Mitigation of Single Event Upset on the Accuracy of Neural Network Accelerators
Huang, Y., Liu, K., Zhu, W., & Zhang, N. (2024) · IEEE Information Technology, Networking, Electronic and Automation Control Conference (ITNEC) · doi:10.1109/itnec60942.2024.10733146 · Cited by 1
“However, the number of parameters and inference computing of a typical DNN is large, which can be 10 6 and 10 9 orders of magnitude, respectively [3].”
BackgroundLow-Complexity Data-Driven Communication Neural Receivers
Wu, Q., Liang, Y., Ng, B. K., Lam, C., & Ma, Y. (2025) · IEEE Access · doi:10.1109/access.2024.3524571 · Cited by 2
“Various neural networks [21], [22], [23] applied TT and TR to compress network to overcoming the curse of dimensionality.”
Quantizing Neural Networks with Knowledge Distillation for Efficient Video Quality Assessment
Yu, J. Y., & Li, Y. (2024) · IEEE International Conference on Visual Communications and Image Processing (VCIP) · doi:10.1109/vcip63160.2024.10849844 · Cited by 1
“Regarding to reduce the size of VQA models, a natural and convenient choice is model quantization, which has been well investigated and shown significant success in efficient model deployment [20]–[24].”
A Data Distribution-Based Ensemble Generation Applied to Wind Speed Forecasting
Almeida, D. M., Neto, P. S. G. D. M., & Cunha, D. C. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-79029-4_2 · Cited by 1
Shah, H. H. (2025) · Global Journal of Emerging AI and Computing · doi:10.70445/gjeac.1.1.2025.59-69 · Cited by 1
Liang, Y., Lam, C., Wu, Q., Ng, B. K., & Im, S. (2025) · IEEE Communications Letters · doi:10.1109/lcomm.2025.3546260 · Cited by 1
“Tensor decomposition, including tensor-ring (TR) decomposition have been widely used in model compression in recent years because they can be used to overcome the curse of dimensionality [2], [3], [4], [5].”
A Sparsity Based Adaptive Compression Algorithm for Model Compression
Lv, Y., & Hu, Y. (2024) · International Conference on Electrical Engineering and Control Technologies (CEECT) · doi:10.1109/ceect63656.2024.10898599 · Cited by 2
“Therefore, model compression has emerged as a critical research direction to achieve high accuracy and high-bandwidth real-time computation in edge computing scenarios [9].”
Approximate Bisimulation Relation Restoration for Neural Networks Based On Knowledge Distillation
Chen, J., Mo, Z., Wang, T., & Xiang, W. (2024) · International Conference on Machine Learning and Applications · doi:10.1109/icmla61862.2024.00137 · Cited by 1
“Mainstream model compression techniques include knowledge distillation to train lightweight model architectures, pruning to remove redundant neurons, and quantization to reduce representation precision [5], [6].”
F2PQNN: a fast and secure two-party inference on quantized convolutional neural networks
Li, J., Yuan, P., Zhang, J., Shen, S., He, Y., & Xiao, R. (2025) · The Computer Journal · doi:10.1093/comjnl/bxaf019 · Cited by 1
Bridging Efficient and Explainable Traffic Flow Prediction on the Edge
Barbareschi, M., Emmanuele, A., Mazzocca, N., & di Torrepadula, F. R. (2025) · Lecture notes on data engineering and communications technologies · doi:10.1007/978-3-031-87778-0_34 · Cited by 1
ChannelZip: SLO-Aware Channel Compression for Task-Adaptive Model Serving on IoT Devices
Luo, P., Hou, J., Tan, H., Yuan, M., Wu, G., Guo, K., Wang, Z., & Li, X. (2025) · ACM Transactions on Sensor Networks · doi:10.1145/3729534 · Cited by 1
“A rich literature has explored many model compression approaches, including the design of lightweight model architecture [3, 11], parameter quantization [2, 22], tensor decomposition [53], and network pruning [4, 24, 36].”
Shen, Y., Wu, B., Straeussnigg, D., & Gutierrez, E. (2025) · Sensors · doi:10.3390/s25082550 · Cited by 2
“In [22] a comprehensive survey on the efficient compression and execution of deep neural networks from both algorithms and hardware perspectives is provided.”
WinAcc: Window-based Acceleration of Neural Networks Using Block Floating Point
Ju, X., He, J., Wen, M., Feng, J., Cao, Y., Shen, J., Chen, Z., & Shi, Y. (2025) · Design, Automation, and Test in Europe · doi:10.23919/date64628.2025.10992780 · Cited by 1
“To address these challenges, various model compression techniques have been proposed [1], [2], with per-vector scaling [3] standing out as a promising solution.”
Afza, N., & Suresh, M. (2025) · International Conference on Knowledge Engineering and Communication Systems (ICKECS) · doi:10.1109/ickecs65700.2025.11035578 · Cited by 1
Xue, J., Liu, Y., & Li, S. (2025) · International Symposium on Computer Technology and Information Science (ISCTIS) · doi:10.1109/isctis65944.2025.11065099 · Cited by 1
MSG: Stealing data from pruned neural networks via malicious sparsity guidance
Shang, J., Wang, J., Wang, K., Jiang, N., & Liu, J. (2025) · Neural Networks · doi:10.1016/j.neunet.2025.108036 · Cited by 1
Gu, J., Liu, C., Chen, Y., Shen, Z., Mao, W., Han, G., He, L., & Xu, J. (2025) · IEEE International System-on-Chip Conference (SOCC) · doi:10.1109/socc66126.2025.11235312 · Cited by 1
“With the advancement of Artificial Intelligence (AI) models, hardware computing platforms are facing unprecedented challenges [1], [2].”
FATE: A Fourier Accelerated Tensor Engine for Neural Networks
Adiletta, J., Islam, B., & Güler, Ü. (2025) · IEEE International Midwest Symposium on Circuits and Systems (MWSCAS) · doi:10.1109/mwscas53549.2025.11244307 · Cited by 1
“Although analog circuits are not as precise compared to digital devices, AI models, particularly neural networks, can tolerate reduced precision in operations and model weights with minimal impact on accuracy [8], [9].”
Nor-Azman, M. N. A., Sheikh, U. U., Mohammed, M. S., Sirkunan, J., & Marsono, M. N. (2025) · Evolutionary Intelligence · doi:10.1007/s12065-025-01122-x · Cited by 1
“The rationale behind layer-wise channel pruning instead of global pruning is that different layers have varying redundancies and should be pruned at different percentages [ 41].”
QuFi: Adaptive Tiled Gustavson Output Reuse for Edge Sparse DNN Accelerators
Navarro, A. F., Cano, J., Abellán, J. L., & Acacio, M. E. (2025) · IEEE International Conference on Computer Design: VLSI in Computers and Processors (ICCD) · doi:10.1109/iccd65941.2025.00023 · Cited by 1
Wei, C., Li, Q., Zhang, A., & Wang, T. (2026) · Brain Sciences · doi:10.3390/brainsci16010083 · Cited by 1
“Unlike traditional deep learning architectures that rely on parameter redundancy to ensure performance [11,14], our design reconciles these conflicting demands through the integration of Ring-GRU and FiLM.”
GEMM-Accelerated ECG Classification via Depthwise Separable 1D-CNN
Hua, T., Lei, L., & Wang, X. (2025) · IEEE Biomedical Circuits and Systems Conference (BioCAS) · doi:10.1109/biocas67066.2025.00024 · Cited by 1
“However, conventional CNN architectures are often associated with high computational complexity and a large number of parameters, making them unsuitable for deployment on power-and resource-constrained edge devices [3].”
A Novel 4-Bit CMOS Based Full Adder for Low-Power IoT and Edge Computing Applications
Vishwakarma, V., Mittal, A., Gupta, B. B., Chui, K. T., & Vishvakarma, S. K. (2025) · International SoC Design Conference · doi:10.1109/isocc66390.2025.11329980 · Cited by 1
Zhang, L., Zhang, F., Mu, D., Tai, Y., Zhang, S., Fan, X., Yang, H., & Liu, Y. (2026) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2026.3656474 · Cited by 1
Federated neuro-symbolic rule learning for lightweight smart building operations
Farrukh, F. F., & Amayri, M. (2026) · Energy and Buildings · doi:10.1016/j.enbuild.2026.117025 · Cited by 3
Chen, Y., Dai, W., Qin, Z., & Han, J. (2026) · Applied Soft Computing · doi:10.1016/j.asoc.2026.114712 · Cited by 1
EqBaB: Efficient equivalence verification for compressed DNNs with bound propagation
Mo, Z., & Xiang, W. (2026) · Neurocomputing · doi:10.1016/j.neucom.2026.132791 · Cited by 1
Zhao, J., Jia, S., Di, J., Lian, J., Ma, Y., Xin, Y., Dang, J., & Liu, J. (2026) · Infrared Physics & Technology · doi:10.1016/j.infrared.2026.106417 · Cited by 1
Evaluation of CNN Models and Benchmark to the Hardware Accelerator
Poornashree, S. J., Joshi, P. V., Sudharshan, K. M., & George, T. M. (2026) · International Conference on Intelligent and Innovative Technologies in Computing, Electrical and Electronics (IITCEE) · doi:10.1109/iitcee67948.2026.11394409 · Cited by 1
“Deng et al. [7] conducted a comprehensive survey on model compression and hardware acceleration, discussing methods like pruning and quantization to optimize neural network models for deployment on embedded devices.”
Yan, S., Shao, H., Deng, X., Jiang, Y., Cheng, K., Zhang, M., & Wang, F. (2026) · Knowledge-Based Systems · doi:10.1016/j.knosys.2026.115633 · Cited by 2
Wang, Y., Wang, M., Shi, J., Ni, Z., & Li, G. (2026) · Frontiers in Marine Science · doi:10.3389/fmars.2026.1778827 · Cited by 2
Self-citationNoise-resilient photonic neural networks through adaptive quantization
Paolini, E., Marinis, L. D., Kincaid, P. S., Valcarenghi, L., Contestabile, G., Roumpos, I., Moralis‐Pegios, M., Pleros, N., & Andriolli, N. (2026) · Neural Computing and Applications · doi:10.1007/s00521-026-11984-6 · Cited by 1
Model pruning method for differentiated image classification tasks on edge devices
Wang, Y., Shi, B., Zhang, H., Xie, Z., & Huang, J. (2026) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2026.103782 · Cited by 1
DustNET: Enabling machine learning and AI models of dusty plasmas
Wang, Z. (. A., Burton, J., Dormagen, N., Du, C.-R., Feng, Y., Foster, J. E., Glenn, S., Klein, M., Knapek, C. A., Matthews, L., et al. (2026) · Physics of Plasmas · doi:10.1063/5.0337599 · Cited by 1
On DNN Model Partitioning and Deployment for Collaborative Multi-Edge AI Computing
Chu, W., Li, Z., Yu, Z., & Lui, J. C. S. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3682802 · Cited by 1
“…techniques that produce a compact model for inference, i.e., through tensor decomposition [23], network sparsification [31] and data quantization [12]; and 2) leveraging the semantic/hierarchical splitting mechanisms that decompose a large layer into components that can be executed in parallel.”
Lin, Y., Zhang, Z., Li, X., Huang, H., Mai, H., Zhan, R., Hu, X., Cai, S., & Xiong, X. (2026) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2026.3683559 · Cited by 1
“grow, creating an urgent demand for e ffi cient model compression techniques [4], [5], [6], [35], [36], [37] and corresponding hardware accelerators.”
Chen, J., & Zou, Y. (2026) · Applied Optics · doi:10.1364/ao.590138 · Cited by 1
Trainable neuromorphic spintronic hardware Via analog finite-difference gradient methods
Pereira, C., Jenkins, A., Raimondo, E., Carpentieri, M., Iranmehr, E., Benetti, L., Roy, S., Ferreira, R., Ventura, J., Finocchio, G., et al. (2026) · Nature Communications · doi:10.1038/s41467-026-73624-w · Cited by 2
“With the limitations of the von Neumann architecture, this translates into severe energy consumption and latency 5 – 7 .”
Implementasi Model MobileNetV2 pada ESP-32 CAM untuk Klasifikasi Botol Plastik
Anjasmara, E. R., Fattah, F., & Gaffar, A. W. M. (2026) · Jurnal Minfo Polgan · doi:10.33395/jmp.v15i2.16197 · Cited by 1
Xu, H., Yang, Z., & Wang, H. (2026) · Applied Sciences · doi:10.3390/app16125766 · Cited by 1
How Low Can Medical Llms Go? an Empirical Study of Low-Bit Quantization
Tasfia, T., Masum, A. K. M., Meyer, E., Cumbo, F., & Aygün, S. (2026) · IEEE Dallas Circuits and Systems Conference (DCAS) · doi:10.1109/dcas69364.2026.11545012 · Cited by 1
“…precision from 16-or 32-bit floating-point representations to lower-bit formats such as 8-bit, 4-bit, 3-bit, or even 2-bit, quantization significantly decreases memory footprint, computational cost, and energy consumption while enabling faster inference and deployment on hardware platforms [26].”
Lightweight High-Performance CNN Design for Mobile Robotic Vision Systems
Afxentiou, V., & Vladimirova, T. (2026) · IEEE Access · doi:10.1109/access.2026.3704908 · Cited by 1
“Quantization reduces the number of bits used to represent each weight or activation, instead of removing parameters, as in pruning [27].”
Efficient Spike-Driven Transformer for High-Performance Drone-View Geo-Localization
Chen, Z., Rong, H.-J., Xu, L., Lu, J., Zou, Z., Yang, Z.-X., & Li, G. (2026) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2026.3716422 · Cited by 3
“These properties can be translated into notable hardware-level energy e ffi ciency [18].”
Self-citationNeuromorphic Object Detection: An In-Depth Study and Future Directions
Li, J., Li, D., Glover, A., Xiaopeng Fan, B., Li, G., Bartolozzi, C., Benosman, R. B., & Tian, Y. (2026) · Proceedings of the IEEE · doi:10.1109/jproc.2026.3717677 · Cited by 1
“However, there has been limited exploration of network compression techniques such as quantization [246], pruning, knowledge distillation [247], and neural architecture search [248], particularly for deployment on dedicated neuromorphic hardware accelerators.”
Self-citationWang, F., Liu, T., Wang, Y., Zhang, Y., Wei, Z., & He, H. (2026) · Electronics · doi:10.3390/electronics15153438 · Cited by 2
Filho, D. S. M. P., Costa, D. M., & de Macedo, D. D. J. (2026) · Future Generation Computer Systems · doi:10.1016/j.future.2026.108738 · Cited by 1
Shit, R. C. (2026) · Computer Communications · doi:10.1016/j.comcom.2026.108654 · Cited by 1
Lasheras, F. S. (2026) · Mathematical Modelling of Natural Phenomena · doi:10.1051/mmnp/2026018 · Cited by 1
HSC: A Hybrid Spin/CMOS Logic Based In-Memory Engine with Area-Efficient Mapping Strategy
Huang, Y., Deng, E., Bai, J., Yang, Q. X., Kang, W., & Pan, B. (2021) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas51556.2021.9401356 · Cited by 1
He-qun, Y. A. N. G., Lu, T., Guo, W., Chang, S., Liu, G., & Luo, Y. (2021) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn52387.2021.9534041 · Cited by 1
“Moreover, in some model compression and acceleration methods, due to the inability to achieve fine granularity control, some valid parameters are also eliminated, or low bits are used instead of high bits for storage, which results in a large loss of model accuracy[14].”
BackgroundDep-$$L_0$$: Improving $$L_0$$-Based Network Sparsification via Dependency Modeling
Li, Y., & Ji, S. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-86523-8_11 · Cited by 1
Wang, Y. (2021) · Computational Intelligence and Neuroscience · doi:10.1155/2021/2391477 · Cited by 1
“In addition, the reuse rate of digital garden planning and design scheme is low, partly because the document storage is not unified and partly because the garden planning and design needs to be based on the actual environment and cannot simply copy the previous excellent planning and design scheme, which requires a lo…”
BackgroundRetractedZhang, Q., Chong, C. W., Abdullah, A. R., & Ali, M. H. (2021) · Computational Intelligence and Neuroscience · doi:10.1155/2021/1370180 · Cited by 1
“studied the optimization of multi-cycle and multi-distribution center [20].”
BackgroundRetractedReal-Time Instance Segmentation for Low-Cost Mobile Robot Systems Based on Computation Offloading
Xie, Y., Guo, Y., Chen, Y., & Mi, Z. (2021) · International Conference on Communications, Computing, Cybersecurity, and Informatics (CCCI) · doi:10.1109/ccci52664.2021.9583186 · Cited by 1
Self-citationAlgorithm/architecture solutions to improve beyond uniform quantization in embedded DNN accelerators
Pedram, A., Ardestani, A. S., Li, L., Abdel-Aziz, H. A., Fang, J., & Hassoun, J. (2022) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2022.102454 · Cited by 1
BenQ: Benchmarking Automated Quantization on Deep Neural Network Accelerators
Wei, Z., Zhang, X., Li, J., Ji, Z., & Wei, J. (2022) · Design, Automation, and Test in Europe · doi:10.23919/date54114.2022.9774652 · Cited by 1
Kernel Modulation: A Parameter-Efficient Method for Training Convolutional Neural Networks
Hu, Y., & Liu, S. (2022) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr56361.2022.9956386 · Cited by 1
“Methods to produce smaller networks include compression of large trained networks (comprehensive review in [3]) and design of run-time efficient models such as ShuffleNetV2 [4], MobileNetV3 [5], and EfficientDet [6].”
MethodOnline Learning for Network Resource Allocation
Salem, T. S. (2022) · ACM SIGMETRICS Performance Evaluation Review · doi:10.1145/3579342.3579348 · Cited by 1
Bhalgaonkar, S., Munot, M. V., & Anuse, A. D. (2022) · International Conference on Signal and Information Processing · doi:10.1109/iconsip49665.2022.10007461 · Cited by 1
“Hence, it becomes essential to compress these models to reduce hyper parameters, memory size and increase inference time while retaining its performance [7].”
BackgroundExplainable Network Pruning for Model Acceleration Based on Filter Similarity and Importance
Wu, J., Nguyen, S., & Alahakoon, D. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-25825-1_16 · Cited by 1
Irregular convolution strategy based tensorized type-2 single layer feedforward network
Li, J., Zhao, G., Huang, S., & Weng, Z. (2023) · International Journal of Machine Learning and Cybernetics · doi:10.1007/s13042-023-01825-6 · Cited by 1
“…to name a few, green supplier selection [11], medical applications, speech recognition, and natural language processing, model compression and hardware acceleration [12, 13], cloud computing environment [14], different tasks evaluation including object detection [15] and semantic segmentation [16].”
BackgroundResultFPGA-Based Accelerator for Rank-Enhanced and Highly-Pruned Block-Circulant Neural Networks
Song, H., Yoon, J., Kim, D., Kwon, E., Oh, T.-H., & Kang, S. (2023) · Design, Automation, and Test in Europe · doi:10.23919/date56975.2023.10137111 · Cited by 1
“Therefore, numerous methods have been proposed to compress and accelerate DNNs [2].”
MethodReal-Time Unsupervised Object Localization on the Edge for Airport Video Surveillance
Ruiz-Barroso, P., Castro, F. M., & Guil, N. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-36616-1_37 · Cited by 3
Metadata
- Authors
- Deng, B. L., Li, G., Han, S., Shi, L., & Xie, Y.
- Affiliations
- Tsinghua University (CN)
- Massachusetts Institute of Technology (US)
- University of California, Santa Barbara (US)
- Venue
- Proceedings of the IEEE
- Published
- 2020-04
- Volume
- 108
- Issue
- 4
- Pages
- 485-532
- ISSN
- 0018-9219 (ISSN-L)
- 1558-2256
- OpenAlex
- W3012561096
- Cited by
- 996
- References
- 334
- Citations by year
- 2026: 131 · 2025: 171 · 2024: 193 · 2023: 174 · 2022: 152 · 2021: 142 · 2020: 31
- FWCI
- 46.83
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- Proceedings of the IEEE (publishedVersion)
- Rare & Special e-Zone (The Hong Kong University of Science and Technology) (submittedVersion)
- Funders
- National Science Foundation (1725447)
- Beijing Academy of Artificial Intelligence (BAAI), Tsinghua University Initiative Scientific Research Program
- Institute for Guo Qiang, Tsinghua University
- 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
- neural network compression, neural network accelerators, deep neural network, data quantization, network sparsification, tensor decomposition, compact modeling, memory efficiency, embedded systems, hardware architecture
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.