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 101–200 of 1,002
Wang, X., Lu, W., Wei, P., Qin, Z., Qiao, N., Qin, X., Zhang, M., Zhu, Y., Bu, L., & Lu, G. (2022) · ACS Applied Materials & Interfaces · doi:10.1021/acsami.2c14625 · Cited by 53
Post-training Quantization for Neural Networks with Provable Guarantees
Zhang, J., Zhou, Y., & Saab, R. (2023) · SIAM Journal on Mathematics of Data Science · doi:10.1137/22m1511709 · Cited by 80
Embedded Deep Learning Accelerators: A Survey on Recent Advances
Akkad, G., Mansour, A., & Inaty, E. N. (2023) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2023.3311776 · Cited by 63
“Moreover, several optimization techniques can be applied to reduce DNNs’ hardware usage, such as pruning [12], quantization [13], [14], knowledge distillation [15], multiplexing [16], and model compression [17], to name a few.”
MethodNeural Architecture Search and Hardware Accelerator Co-Search: A Survey
Sekanina, L. (2021) · IEEE Access · doi:10.1109/access.2021.3126685 · Cited by 60
“Model compression tries to reduce the number of different weight values to minimize the CNN memory footprint [42], [46].”
BackgroundRodríguez, I., Campos, C., & Fdez‐Riverola, F. (2021) · Neural Computing and Applications · doi:10.1007/s00521-021-06830-w · Cited by 60
Real Time Power Equipment Meter Recognition Based on Deep Learning
Fan, Z., Shi, L., Xi, C., Wang, H., Wang, S., & Wu, G. (2022) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2022.3191709 · Cited by 61
Self-citationGenExp: Multi-objective pruning for deep neural network based on genetic algorithm
Xu, K., Zhang, D., An, J., Liu, L., Liu, L., & Wang, D. (2021) · Neurocomputing · doi:10.1016/j.neucom.2021.04.022 · Cited by 55
Global Aligned Structured Sparsity Learning for Efficient Image Super-Resolution
Wang, H., Zhang, Y., Qin, C., Gool, L. V., & Fu, Y. (2023) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2023.3268675 · Cited by 55
“In addition to these attempts, others try to achieve efficient SR by tapping into the advances in model compression [12], [13], [14], [15], [16] 1 –”
BackgroundTransforming Large-Size to Lightweight Deep Neural Networks for IoT Applications
Mishra, R., & Gupta, H. P. (2023) · ACM Computing Surveys · doi:10.1145/3570955 · Cited by 66
“…covered the shortcomings of including layer compression but cannot summarize the vital technique of knowledge distillation for improving the performance of compressed DNNs. Similarly, Deng et al. [40] covered the DNN compression techniques emphasizing on co-designing of the hardware and algorithm.”
BackgroundEmbedded Real-Time Vehicle and Pedestrian Detection Using a Compressed Tiny YOLO v3 Architecture
Falaschetti, L., Manoni, L., Palma, L., Pierleoni, P., & Turchetti, C. (2024) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2024.3447453 · Cited by 56
“Several different methods, such as pruning [84], [85], [86], [87], [88], quantization [89] and tensor decomposition are available for this purpose.”
DeepAbstract: Neural Network Abstraction for Accelerating Verification
Ashok, P., Hashemi, V., Křetínský, J., & Mohr, S. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-59152-6_5 · Cited by 50
Distributed Machine Learning in Edge Computing: Challenges, Solutions and Future Directions
Tu, J., Yang, L., & Cao, J. (2025) · ACM Computing Surveys · doi:10.1145/3708495 · Cited by 57
“The specific methods for low-dimensional tensor are SVD and full-rank decomposition, and for high-dimensional tensor are Tucker decomposition and CP decomposition, and so on [28, 185].”
Hybrid tensor decomposition in neural network compression
Wu, B., Wang, D., Zhao, G., Deng, L., & Li, G. (2020) · Neural Networks · doi:10.1016/j.neunet.2020.09.006 · Cited by 48
Self-citationDigital Retina: A Way to Make the City Brain More Efficient by Visual Coding
Gao, W., Ma, S., Duan, L., Tian, Y., Xing, P., Wang, Y., Wang, S., Jia, H., & Huang, T. (2021) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2021.3104305 · Cited by 53
An FPGA-based MobileNet Accelerator Considering Network Structure Characteristics
Yan, S., Liu, Z., Wang, Y., Zeng, C., Liu, Q., Cheng, B., & Cheung, R. C. C. (2021) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl53798.2021.00011 · Cited by 56
Mulo, J., Liang, H., Qian, M., Biswas, M., Rawal, B. S., Guo, Y., & Yu, W. (2025) · Future Internet · doi:10.3390/fi17030107 · Cited by 56
“…such as model pruning, which reduces unnecessary parameters; quantization, which converts models into lower-precision formats; and knowledge distillation, where smaller models learn from larger ones, are key to significantly lowering computational requirements while maintaining accuracy [275,276].”
Optimizing Edge AI: A Comprehensive Survey on Data, Model, and System StrategiesPreprint
Wang, X., & Jia, W. (2025) · Qeios · doi:10.32388/izohch · Cited by 56
Huang, Q. (2022) · AI · doi:10.3390/ai3010011 · Cited by 48
“According to [21,22], the common model compression approaches include the following: network pruning [23], weight factorization [24], quantization [25], and weight sharing [26].”
MethodOptimal Lottery Tickets via SubsetSum: Logarithmic Over-Parameterization is SufficientPreprint
Pensia, A., Rajput, S., Nagle, A., Vishwakarma, H., & Papailiopoulos, D. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.07990 · Cited by 109
“…is, σ ( x ) = x · 1 x 0 . is − A second, larger network g ( x ) of the following form Our goal is to obtain a pruned version of g by eliminating weights, i.e., ˆ g ( x ) = ( S 2 l M 2 l ) σ (( S 2 l 1 M 2 l 1 ) . . . σ (( S 1 M 1 ) x )) , where each S i is a binary (pruning) matrix, with the same…”
BackgroundMethodInfluentialFalaschetti, L., Manoni, L., Rivera, R. C. F., Pau, D., Romanazzi, G., Silvestroni, O., Tomaselli, V., & Turchetti, C. (2021) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2021.3098454 · Cited by 45
Multiple hierarchical compression for deep neural network toward intelligent bearing fault diagnosis
Sun, J., Liu, Z., Wen, J., & Fu, R. (2022) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2022.105498 · Cited by 46
TailorFL: Dual-Personalized Federated Learning under System and Data Heterogeneity
Deng, Y., Chen, W., Ren, J., Lyu, F., Liu, Y., Liu, Y., & Zhang, Y. (2022) · ACM Conference on Embedded Networked Sensor Systems (SenSys) · doi:10.1145/3560905.3568503 · Cited by 57
“In contrast, to obtain submodels with reduced overheads, model pruning is a suitable choice [8].”
BackgroundDayal, A., Paluru, N., Cenkeramaddi, L. R., Soumya, J., & Yalavarthy, P. K. (2021) · Electronics · doi:10.3390/electronics10020182 · Cited by 46
“The state-of-the-art performance has been achieved with the help of complex models that require more computational resources [25].”
BackgroundInfluentialNeural Network Structure Optimization by Simulated Annealing
Kuo, C.-L., Kuruoğlu, E. E., & Chan, W. K. (2022) · Entropy · doi:10.3390/e24030348 · Cited by 48
“In order to actually reduce the total size of a model while doing inference in a mobile electronic device, various references found in the survey [21] focus on pruning the weight parameters directly instead of the gradients.”
BackgroundSkydiver: A Spiking Neural Network Accelerator Exploiting Spatio-Temporal Workload Balance
Chen, Q., Gao, C., Fang, X., & Luan, H. (2022) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3158834 · Cited by 44
“To deal with this problem, many researchers designed DNN compression methods and efficient hardware architectures [1].”
MethodThe Future of Large Language Model Pre-training is FederatedPreprint
Sani, L., Iacob, A., Cao, Z., Marino, B., Gao, Y., Paulik, T., Zhao, W., Shen, W. F., Aleksandrov, P., Qiu, X., et al. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2405.10853 · Cited by 59
“Before sending these updates ( L. 27 ), clients apply local post-processing steps such as gradient clipping and noise injection in the case of differential privacy [48, 78], or model compression [79] for lower server-client communication costs.”
BackgroundMethodLuo, H., Chen, T., Li, X., Li, S., Zhang, C., Zhao, G., & Liu, X. (2022) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2022.3157957 · Cited by 43
“Model compression has been commonly adopted to effi-ciently alleviate the tension between resource-hungry DNN and resource-constrained edge devices [18].”
MethodDistributed computing in multi-agent systems: a survey of decentralized machine learning approaches
Ahmed, I., Syed, M. A., Maaruf, M., & Khalid, M. (2024) · Computing · doi:10.1007/s00607-024-01356-0 · Cited by 55
FxHENN: FPGA-based acceleration framework for homomorphic encrypted CNN inference
Zhu, Y., Wang, X., Ju, L., & Guo, S. (2023) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca56546.2023.10071133 · Cited by 42
Ruan, D., Han, J., Yan, J., & Gühmann, C. (2023) · Scientific Reports · doi:10.1038/s41598-023-31532-9 · Cited by 42
“Works presented in 24 and 25 use an iterative threshold-based pruning to learn the weights and simultaneously prune the unimportant ones.”
BackgroundImtiaz, N., Wahid, A., Abideen, S. Z. U., Kamal, M. M., Sehito, N., Khan, S. U., Virdee, B. S., Kouhalvandi, L., & Alibakhshikenari, M. (2025) · Photonics · doi:10.3390/photonics12010035 · Cited by 56
Wang, R., Bai, Q., Ao, J., Zhou, L., Xiong, Z., Wei, Z., Zhang, Y., Ko, T., & Li, H. (2022) · Interspeech 2022 · doi:10.21437/interspeech.2022-10269 · Cited by 40
Low-power deep learning edge computing platform for resource constrained lightweight compact UAVs
Albanese, A., Nardello, M., & Brunelli, D. (2022) · Sustainable Computing Informatics and Systems · doi:10.1016/j.suscom.2022.100725 · Cited by 60
Reggiani, E., Pappalardo, A., Doblas, M., Moretó, M., Olivieri, M., Ünsal, O., & Cristal, A. (2023) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca56546.2023.10071076 · Cited by 54
“Accelerating quantized DNNs represents a widespread research topic [16], [20], [30].”
BackgroundS-RAN: Semantic-Aware Radio Access Networks
Sun, Y., Zhang, L., Guo, L., Li, J., Niyato, D. T., & Fang, Y. M. (2024) · IEEE Communications Magazine · doi:10.1109/mcom.004.2400105 · Cited by 40
“Recent research has introduced hardware-efficient ML techniques [10], shedding light on efficient S-JSCC de-sign.”
BackgroundMethodYuan, G., Liao, Z., Ma, X., Cai, Y., Kong, Z., Shen, X., Fu, J., Li, Z., Zhang, C., Peng, H., et al. (2021) · International Symposium on Quality Electronic Design (ISQED) · doi:10.1109/isqed51717.2021.9424332 · Cited by 39
“DNN Weight pruning [8], [28]–[31], as one of the DNN model compression techniques, has also been investigated to reduce the weight storage and improve performance for ReRAM-based acceleration designs [32], [33].”
MethodSurvey of Deep Learning Accelerators for Edge and Emerging Computing
Alam, S., Yakopcic, C., Wu, Q., Barnell, M., Khan, S., & Taha, T. M. (2024) · Electronics · doi:10.3390/electronics13152988 · Cited by 38
Shao, X., Chen, H., Magson, K., Wang, J., Song, J., Chen, J., & Sasaki, J. (2024) · Aquatic Conservation Marine and Freshwater Ecosystems · doi:10.1002/aqc.4241 · Cited by 41
“A good model should achieve an optimal trade-off between these two criteria (Deng et al., 2020; Götz et al., 2022).”
MethodA Holistic Review of the TinyML Stack for Predictive Maintenance
Njor, E., Hasanpour, M. A., Madsen, J., & Fafoutis, X. (2024) · IEEE Access · doi:10.1109/access.2024.3512860 · Cited by 53
Performance enhancement of artificial intelligence: A survey
Krichen, M., & Abdalzaher, M. S. (2024) · Journal of Network and Computer Applications · doi:10.1016/j.jnca.2024.104034 · Cited by 41
Ning, H., Wen, H., Meng, Y., Yu, Z., Fu, Y., Zou, X., Shen, Y., Luo, X., Zhao, Q., Zhang, T., et al. (2025) · Nature Electronics · doi:10.1038/s41928-024-01328-4 · Cited by 40
“It achieves the highest training accuracy among all evaluated granularities and may outperform the dense model in certain configurations 10,11 .”
InfluentialRecent Progress on Memristive Convolutional Neural Networks for Edge Intelligence
Qin, Y., Bao, H., Wang, F., Chen, J., Li, Y., & Miao, X. (2020) · Advanced Intelligent Systems · doi:10.1002/aisy.202000114 · Cited by 40
Reconfigurability, Why It Matters in AI Tasks Processing: A Survey of Reconfigurable AI Chips
Wei, S., Lin, X., Tu, F., Wang, Y., Liu, L., & Yin, S. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3228860 · Cited by 40
“But they often meets more difficulties in maintaining the model accuracy [12].”
BackgroundNonlinear tensor train format for deep neural network compression
Wang, D., Zhao, G., Chen, H., Liu, Z., Deng, L., & Li, G. (2021) · Neural Networks · doi:10.1016/j.neunet.2021.08.028 · Cited by 38
Self-citationHeterogeneous Systolic Array Architecture for Compact CNNs Hardware Accelerators
Xu, R., Ma, S., Wang, Y., Guo, Y., Li, D., & Qiao, Y. (2021) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2021.3129647 · Cited by 43
“A popular choice is the depthwise convolution (DWConv) [1], [2], [3].”
BackgroundMethodInfluentialA Method of Deep Learning Model Optimization for Image Classification on Edge Device
Lee, H., Lee, N., & Lee, S. (2022) · Sensors · doi:10.3390/s22197344 · Cited by 42
“Due to model optimization and network compression, the memory size and a computational delay of deep learning models are reduced compared to the native neural networks, while the performance of models is maintained as well [7].”
Can Unstructured Pruning Reduce the Depth in Deep Neural Networks?
Liao, Z., Quétu, V., Nguyen, V.-T., & Tartaglione, E. (2023) · IEEE International Conference on Computer Vision Workshops · doi:10.1109/iccvw60793.2023.00151 · Cited by 54
“To address these issues, researchers have developed various techniques to reduce the size of neural networks without sacrificing their performance, including quantization [9, 26], compression [6, 24], and pruning [10, 11, 15, 16, 23].”
MethodHohman, F., Kery, M. B., Ren, D., & Moritz, D. (2024) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3613904.3642109 · Cited by 50
“For an in-depth review of the technical descriptions of compression techniques, see the following surveys: [17, 19, 21].”
BackgroundA Lightweight Deep Learning Algorithm for WiFi-Based Identity Recognition
Cao, Y., Zhou, Z., Zhu, C., Duan, P., Chen, X., & Li, J. (2021) · IEEE Internet of Things Journal · doi:10.1109/jiot.2021.3078782 · Cited by 36
Architectural Vision for Quantum Computing in the Edge-Cloud Continuum
Furutanpey, A., Barzen, J., Bechtold, M., Dustdar, S., Leymann, F., Raith, P., & Truger, F. (2023) · IEEE International Conference on Quantum Software (QSW) · doi:10.1109/qsw59989.2023.00021 · Cited by 40
“, IoT devices) or DNNs compressed with quantization, pruning, or knowledge distillation methods [16], [26] as “lightweight”.”
MethodMemristor-based adaptive analog-to-digital conversion for efficient and accurate compute-in-memory
Hong, H., Du, Z., Jiang, M., Mao, R., Ren, Y., Li, F., Mao, W., Peng, M., Zhang, W., Liu, Z., et al. (2025) · Nature Communications · doi:10.1038/s41467-025-65233-w · Cited by 37
Secure XOR-CIM Engine: Compute-In-Memory SRAM Architecture With Embedded XOR Encryption
Huang, S., Jiang, H., Peng, X., Li, W., & Yu, S. (2021) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2021.3120296 · Cited by 36
Huang, X., Zhang, D., Hu, X., Ye, C., & Zhang, K. (2021) · Journal of Lightwave Technology · doi:10.1109/jlt.2021.3128579 · Cited by 34
Zheng, Y., Sun, P., Ren, Q., Xu, W., & Zhu, D. (2023) · Neurocomputing · doi:10.1016/j.neucom.2023.127124 · Cited by 38
Forest Fire Detection Based on Lightweight Yolo
Wang, S., Chen, T., Lv, X., Zhao, J., Zou, X., Zhao, X., Xiao, M., & Wei, H. (2021) · Chinese Control and Decision Conference (CCDC) · doi:10.1109/ccdc52312.2021.9601362 · Cited by 38
QTTNet: Quantized tensor train neural networks for 3D object and video recognition
Lee, D., Wang, D., Yang, Y., Deng, L., Zhao, G., & Li, G. (2021) · Neural Networks · doi:10.1016/j.neunet.2021.05.034 · Cited by 36
Self-citationEfficient Environmental Context Prediction for Lower Limb Prostheses
Zhong, B., da Silva, R. L., Tran, M., Huang, H., & Lobaton, E. J. (2021) · IEEE Transactions on Systems Man and Cybernetics Systems · doi:10.1109/tsmc.2021.3084036 · Cited by 33
“progress has been made toward neural network compression and deep learning specific hardware design [18].”
BackgroundPeaTMOSS: A Dataset and Initial Analysis of Pre-Trained Models in Open-Source Software
Jiang, W., Yasmin, J., Jones, J., Synovic, N. M., Kuo, J., Bielanski, N., Tian, Y., Thiruvathukal, G. K., & Davis, J. C. (2024) · International Conference on Mining Software Repositories (MSR) · doi:10.1145/3643991.3644907 · Cited by 49
“Furthermore, models can be optimized for size and efficiency to run on edge devices, a process known as model compression [27].”
A Survey on Hardware Accelerator Design of Deep Learning for Edge Devices
Samanta, A., Hatai, I., & Mal, A. K. (2024) · Wireless Personal Communications · doi:10.1007/s11277-024-11443-2 · Cited by 39
“The processing in memory (PIM) [54] technique is used to reduce data travel while enhancing MVM (matrix–vector-multiplication) efficiency.”
Qu, X., Aponte, D., Banbury, C., Robinson, D. P., Ding, T., Koishida, K., Zharkov, I., & Chen, T. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.01419 · Cited by 55
“Quantization reduces the bit width of the data flowing through a DNN [14].”
DFE: efficient IoT network intrusion detection using deep feature extraction
Basati, A., & Faghih, M. M. (2022) · Neural Computing and Applications · doi:10.1007/s00521-021-06826-6 · Cited by 34
“Some methods were recently used to lower the computational complexity of a deep model, including convolution factorization [18] and network compression [19].”
MethodA Serverless Computing Fabric for Edge & Cloud
Nastić, S., Raith, P., Furutanpey, A., Pusztai, T. W., & Dustdar, S. (2022) · IEEE International Conference on Cognitive Machine Intelligence (CogMI) · doi:10.1109/cogmi56440.2022.00011 · Cited by 46
“The systematic comparison of existing approaches focuses on implementations, efficiency, scheduling, benchmarks, security aspects, and source code availability.”
Fresa, A., & Champati, J. P. (2023) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2023.3267458 · Cited by 32
“The objectives that were considered for optimizing the offloading decision are, 1) minimize job execution times, see for example [17], [18], [19], [20], 2) minimize the energy of the ED spent in computing and/or transmitting the jobs, subject to a constraint on the execution delay, see for example [21], [22], [23] and…”
MethodInfluentialReal-Time Semantic Segmentation: A brief survey and comparative study in remote sensing
Broni-Bediako, C., Xia, J., & Yokoya, N. (2023) · IEEE Geoscience and Remote Sensing Magazine · doi:10.1109/mgrs.2023.3321258 · Cited by 33
“Several handcrafted convolutional building blocks have been proposed to reduce the number of FLOPs in convolution operations, and are widely used to develop models for efficient inference [48, 62].”
MethodTraining-free neural architecture search: A review
Wu, M., & Tsai, C. (2023) · ICT Express · doi:10.1016/j.icte.2023.11.001 · Cited by 36
Zhu, M., Gong, Y., Tian, C., & Zhu, Z. (2024) · Drones · doi:10.3390/drones8080412 · Cited by 33
“In order to adapt to the resource constraints of edge devices, researchers have proposed a variety of model compression and optimization techniques, including pruning, quantization, and knowledge distillation [103].”
BackgroundDependable DNN Accelerator for Safety-Critical Systems: A Review on the Aging Perspective
Moghaddasi, I., Gorgin, S., & Lee, J. (2023) · IEEE Access · doi:10.1109/access.2023.3300376 · Cited by 30
Deep Learning Model Compression With Rank Reduction in Tensor Decomposition
Dai, W., Fan, J., Miao, Y., & Hwang, K. (2023) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2023.3330542 · Cited by 42
Kronecker CP Decomposition With Fast Multiplication for Compressing RNNs
Wang, D., Wu, B., Zhao, G., Yao, M., Chen, H., Deng, L., Yan, T., & Li, G. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3105961 · Cited by 37
“Therefore, many researchers have aimed their attention to compressing RNNs [7].”
BackgroundSelf-citationQLP: Deep Q-Learning for Pruning Deep Neural Networks
Camci, E., Gupta, M., Wu, M., & Lin, J. (2022) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2022.3167951 · Cited by 32
“Various compression techniques [5], [6] have been pro-41 posed to alleviate these issues, including pruning [7], weight sharing [8], quantization [9], knowledge distillation [10], tensor factorisation & decomposition [11], and architecture search [12].”
MethodFurutanpey, A., Raith, P., & Dustdar, S. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3381952 · Cited by 45
“Nevertheless, they are bound to reduced predictive strength relative to their contemporary unconstrained counterparts [34].”
Li, J., Magar, R. T., Chen, D., Lin, F., Wang, D., Yin, X., Zhuang, W., & Li, Z. (2024) · Computers and Electronics in Agriculture · doi:10.1016/j.compag.2024.108861 · Cited by 35
Compressing 3DCNNs based on tensor train decomposition
Wang, D., Zhao, G., Li, G., Deng, L., & Wu, Y. (2020) · Neural Networks · doi:10.1016/j.neunet.2020.07.028 · Cited by 33
Self-citationNa, J., Zhang, H., Lian, J., & Zhang, B. (2022) · Applied Sciences · doi:10.3390/app122010619 · Cited by 30
Multiobjective Evolutionary Generative Adversarial Network Compression for Image Translation
Zhou, Y., Hu, B., Yuan, X., Huang, K., Zhang, Y., & Yen, G. G. (2023) · IEEE Transactions on Evolutionary Computation · doi:10.1109/tevc.2023.3261135 · Cited by 30
“To reduce the computational complexity of neural network models, model compression techniques [11], including filter pruning [12], network slimming [13], or knowledge distillation (KD) [14], have been widely investigated.”
BackgroundMethodCompiler Technologies in Deep Learning Co-Design: A Survey
Zhang, H., Xing, M., Wu, Y., & Zhao, C. (2023) · Intelligent Computing · doi:10.34133/icomputing.0040 · Cited by 41
Liu, Z., Liu, Q., Yan, S., & Cheung, R. C. C. (2023) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3615661 · Cited by 29
“To meet the requirements in real-time processing and memory resources in embedded systems, there are growing interests in neural network compression and dedicated hardware accelerators [4].”
BackgroundBlockchain-Based Clustered Federated Learning for Non-Intrusive Load Monitoring
Wang, T., & Dong, Z. Y. (2023) · IEEE Transactions on Smart Grid · doi:10.1109/tsg.2023.3326194 · Cited by 33
“The data quantization is commonly achieved by projection, and the projection function is as follows [38]: where z denotes the initial value in a continuous space; φ( z ) is the quantized value in a discrete space; d is the quantization step length.”
BackgroundLarge Language Models in Psychiatry: Current Applications, Limitations, and Future Scope
Liu, Z., Bao, Y., Zeng, S., Qian, R., Deng, M., Gu, A., Li, J., Wang, W., Cai, W., Li, W., et al. (2024) · Big Data Mining and Analytics · doi:10.26599/bdma.2024.9020046 · Cited by 31
An Efficient Deep Neural Network for Surface Defect Detection in Industrial Edge Sensing
Wang, J., Zou, H., Zhou, M., & Su, R. (2024) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2024.3507954 · Cited by 30
“So there is a growing realization about the lightweight and efficient improvement [13], [14].”
2D Van Der Waals Ferroelectric Materials and Devices for Neuromorphic Computing
Wen, Z., Chen, J., Zhang, Q., Wang, G., Wang, X., Yang, F., Liu, F., Luo, X. D., & Liu, F. (2025) · Small · doi:10.1002/smll.202412761 · Cited by 27
The Bitlet Model: A Parameterized Analytical Model to Compare PIM and CPU Systems
Ronen, R., Eliahu, A., Leitersdorf, O., Peled, N., Korgaonkar, K., Chattopadhyay, A., Perach, B., & Kvatinsky, S. (2022) · ACM Journal on Emerging Technologies in Computing Systems · doi:10.1145/3465371 · Cited by 27
“PIM solutions have recently been integrated into application-specific [9] and generalpurpose [19] architectures.”
BackgroundDNN Model Compression for IoT Domain-Specific Hardware Accelerators
Russo, E., Palesi, M., Monteleone, S., Patti, D., Mineo, A., Ascia, G., & Catania, V. (2021) · IEEE Internet of Things Journal · doi:10.1109/jiot.2021.3111723 · Cited by 27
Multi-Camera-Based Sorting System for Surface Defects of Apples
Lee, J.-H., Trong, V. H., Kwon, G.-J., Kim, H., & Kim, J. Y. (2023) · Sensors · doi:10.3390/s23083968 · Cited by 27
“As the size of models increases, various research efforts have emerged to make them usable on limited hardware resources [15].”
BackgroundAnalog Multiply-Accumulate Cell With Multi-Bit Resolution for All-Analog AI Inference Accelerators
Nägele, R., Finkbeiner, J., Stadtlander, V., Grözing, M., & Berroth, M. (2023) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2023.3268728 · Cited by 27
4-bit CNN Quantization Method With Compact LUT-Based Multiplier Implementation on FPGA
Zhao, B., Wang, Y., Zhang, H., Zhang, J., Chen, Y., & Yang, Y. (2023) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2023.3324357 · Cited by 30
“In general, nonsaturated quantization is considered to maintain a higher model accuracy, since nonsaturated quantization with larger scaling factor makes most of the values (near 0) to have higher resolution, especially in low-precision quantization [9], [11], [26].”
BackgroundMethodLuo, X., Liu, D., Kong, H., Huai, S., Chen, H., Xiong, G., & Liu, W. (2024) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3701728 · Cited by 33
From Algorithm to Hardware: A Survey on Efficient and Safe Deployment of Deep Neural Networks
Geng, X., Wang, Z., Chen, C., Xu, Q., Xu, K., Jin, C., Gupta, M., Yang, X., Chen, Z., Sabry Aly, M. M., et al. (2025) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2024.3394494 · Cited by 26
“For example, there is a growing trend to combine KD with quantization and pruning.”
A survey of energy-efficient strategies for federated learning inmobile edge computing
Yan, K., Shu, N., Wu, T., Liu, C., & Yang, P. (2024) · Frontiers of Information Technology & Electronic Engineering · doi:10.1631/fitee.2300181 · Cited by 30
A Decade of Video Analytics at Edge: Training, Deployment, Orchestration, and Platforms
Bai, T., Zhao, H., Huang, L., Wang, Z., Kim, D. I., & Nallanathan, A. (2025) · IEEE Communications Surveys & Tutorials · doi:10.1109/comst.2025.3563377 · Cited by 43
“Specifically, as depicted in Fig 6, weight pruning reduces the number of edges, whereas neuron pruning reduces the number of nodes [98].”
Mobility- and Energy-Aware Cooperative Edge Offloading for Dependent Computation Tasks
Mehrabi, M., Shen, S., Hai, Y., Latzko, V., Koudouridis, G. P., Gelabert, X., Martin Reisslein, B., & Fitzek, F. H. P. (2021) · Network · doi:10.3390/network1020012 · Cited by 29
“The QCQP can be efficiently solved, e.g., with conic optimization solvers [65], especially with accelerators [66,67].”
MethodA Review of Efficient Real-Time Decision Making in the Internet of Things
Kang, K. (2022) · Technologies · doi:10.3390/technologies10010012 · Cited by 32
“…(Convolutional Neural Networks) are created by leveraging the spatial correlation within a convolutional layer to convolve feature maps with multiple weight kernels (Compact RNNs (Recurrent Neural Networks) for sequence data analysis has also received significant attention from researchers [78].).”
BackgroundMethodInfluentialStage-Wise Magnitude-Based Pruning for Recurrent Neural Networks
Li, G., Yang, P., Qian, C., Hong, R., & Tang, K. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3184730 · Cited by 26
“Next, we will briefly introduce non-pruning works for compressing a trained DNN [7], [39].”
MethodSelf-citationMCA-YOLOV5-Light: A Faster, Stronger and Lighter Algorithm for Helmet-Wearing Detection
Sun, C., Zhang, S., Qu, P., Wu, X., Feng, P., Tao, Z., Zhang, J., & Wang, Y. (2022) · Applied Sciences · doi:10.3390/app12199697 · Cited by 26
A Survey on Self-Evolving Autonomous Driving: A Perspective on Data Closed-Loop Technology
Li, X., Wang, Z., Huang, Y., & Chen, H. (2023) · IEEE Transactions on Intelligent Vehicles · doi:10.1109/tiv.2023.3319689 · Cited by 32
“Other works that research model compression and hardware acceleration technologies are reviewed in [91].”
BackgroundReusing Deep Learning Models: Challenges and Directions in Software Engineering
Davis, J. C., Jajal, P., Jiang, W., Schorlemmer, T. R., Synovic, N. M., & Thiruvathukal, G. K. (2023) · IEEE John Vincent Atanasoff International Symposium on Modern Computing (JVA) · doi:10.1109/jva60410.2023.00015 · Cited by 38
“For example, the adaptation of DNNs to embedded devices (also known as IoT or Edge devices) often demands significant engineering efforts, encompassing techniques such as model compression and hardware acceleration [67].”
MethodReal-Time Network Packet Classification Exploiting Computer Vision Architectures
Paolini, E., Valcarenghi, L., Maggiani, L., & Andriolli, N. (2024) · IEEE Open Journal of the Communications Society · doi:10.1109/ojcoms.2024.3363082 · Cited by 26
“However, advancements in hardware acceleration (like specialized chips or GPUs) [45] and optimization techniques [46] have significantly improved their efficiency.”
BackgroundSME: ReRAM-based Sparse-Multiplication-Engine to Squeeze-Out Bit Sparsity of Neural Network
Liu, F., Zhao, W., He, Z., Wang, Z., Zhao, Y., Yang, T., Feng, J., Liang, X., & Jiang, L. (2021) · IEEE International Conference on Computer Design: VLSI in Computers and Processors (ICCD) · doi:10.1109/iccd53106.2021.00072 · Cited by 25
Lall, A., & Tallur, S. G. (2023) · Scientific Reports · doi:10.1038/s41598-023-29277-6 · Cited by 26
“Pruning and quantisation are commonly employed methods for performing model compression 29,30 .”
MethodИгоревич, А. В., Tominov, R. V., Vakulov, Z., Zhavoronkov, L. G., & Smirnov, V. A. (2023) · Nano Research · doi:10.1007/s12274-023-5639-5 · Cited by 25
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.