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 501–600 of 1,002
Urban Fine Management of Multisource Spatial Data Fusion Based on Smart City Construction
Long, Y., Zhang, X., & Feng, G. (2021) · Mathematical Problems in Engineering · doi:10.1155/2021/5058791 · Cited by 6
“Now, in the construction of a smart city based on the multifusion of big data, the research and judgment of traffic roads can better reflect the status of this factor in the fine management of the whole city in the threedimensional space [16, 17].”
BackgroundRetractedEfficient methods for deep learning
Cai, H., Lin, J., & Han, S. (2022) · Elsevier eBooks · doi:10.1016/b978-0-12-822109-9.00013-8 · Cited by 10
Self-citationReLP: Reinforcement Learning Pruning Method Based on Prior Knowledge
Zhang, W., Ji, M., Yu, H., & Zhen, C. (2022) · Neural Processing Letters · doi:10.1007/s11063-022-11058-3 · Cited by 9
“At present, the mainstream model compression methods [4] include compact networks, tensor decomposition, data quantization, knowledge distillation, model pruning, and hybrid methods.”
MethodHMC: Hybrid model compression method based on layer sensitivity grouping
Yang, G., Yu, S., Yang, H., Nie, Z., & Wang, J. (2023) · PLoS ONE · doi:10.1371/journal.pone.0292517 · Cited by 6
“Model compression is the main means to reduce the deep model size [14, 15].”
BackgroundApplication of deep learning in fMRI-based human brain parcellation: a review
Li, Y., Chen, X., Ling, Q., He, Z., & Liu, A. (2023) · Measurement Science and Technology · doi:10.1088/1361-6501/ad14e2 · Cited by 7
Wang, H., Du, J., Dai, Y., Lee, C., Ren, Y., & Liu, Y. (2024) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp48485.2024.10447231 · Cited by 8
“Pruning is an effective method to remove the redundancy in a network, which can be divided into structured pruning and unstructured pruning [17].”
MethodA Novel CNFET SRAM-Based Compute-In-Memory for BNN Considering Chirality and Nanotubes
Kim, Y., Alnatsheh, N., Yadav, N., Cho, J., Jo, H., & Choi, K. K. (2024) · Electronics · doi:10.3390/electronics13112192 · Cited by 6
TinyML using neural networks for resource-constrained devices
Kulkarni, V. Y., & Jujare, V. A. (2024) · Elsevier eBooks · doi:10.1016/b978-0-44-322202-3.00010-5 · Cited by 6
Growing evolutional deep echo state network
Shen, Q., Wang, J., Zhang, H., Peng, J., Sun, M., & Mao, Y. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.128676 · Cited by 6
Toward Real-Time and Efficient Perception Workflows in Software-Defined Vehicles
Sumaiya, Jafarpourmarzouni, R., Luo, Y., Lu, S., & Dong, Z. (2024) · IEEE Internet of Things Journal · doi:10.1109/jiot.2024.3492801 · Cited by 10
“In addition, there are various hardware acceleration techniques [23] like ONNX [24], TensorRT [1], [25], and ONNX Runtime EPI [26], which have been introduced to enhance model performance.”
Distributing Inference Tasks Over Interconnected Systems Through Dynamic DNNs
Singhal, C., Wu, Y., Malandrino, F., Levorato, M., & Chiasserini, C. F. (2025) · IEEE Transactions on Networking · doi:10.1109/ton.2025.3543848 · Cited by 7
“Twomain technical strategies attempt to address this issue: model compression [6] and edge computing [7].”
Towards Optimal Layer Ordering for Efficient Model Compression via Pruning and Quantization
Shinde, T. S. (2025) · IEEE International Conference on Digital Signal Processing (DSP) · doi:10.1109/dsp65409.2025.11074798 · Cited by 6
“Efficient model compression techniques such as pruning [4], [5] and quantization [6], [7] are widely used to reduce model size and inference latency.”
SpecMCTS: Accelerating Monte Carlo Tree Search Using Speculative Tree Traversal
Kim, J., Kang, B., & Cho, H. (2021) · IEEE Access · doi:10.1109/access.2021.3120384 · Cited by 6
“A speculation model can be created in many different ways, such as DNN pruning, quantization, or matrix factorization [15]–[17].”
MethodA New Clustering-Based Technique for the Acceleration of Deep Convolutional Networks
Pikoulis, E. V., Mavrokefalidis, C., Nousias, S., & Lalos, A. S. (2021) · Advances in intelligent systems and computing · doi:10.1007/978-981-16-3357-7_5 · Cited by 7
LNNet: Lightweight Nested Network for motion deblurring
Guo, C., Wang, Q., Dai, H., Wang, H. H., & Li, P. (2022) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2022.102584 · Cited by 6
Wang, Y., Yu, Z., Liu, S., Zhou, Z., & Guo, B. (2023) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3580815 · Cited by 8
“Model compression is a prevailing strategy to reduce the resource overhead of DNNs without notably compromising accuracy [12].”
BackgroundHyBNN: Quantifying and Optimizing Hardware Efficiency of Binary Neural Networks
Yang, G., Lei, J., Fang, Z., Li, Y., Zhang, J., & Xie, W. (2023) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3631610 · Cited by 9
AdaStreamLite: Environment-adaptive Streaming Speech Recognition on Mobile Devices
Wei, Y., Xiong, J., Liu, H., Yu, Y., Pan, J. J., & Du, J. (2023) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3631460 · Cited by 8
“Recently, with the emergence of edge intelligence [71], ubiquitous computing and deep learning communities have been exploring how to transfer the training and inference processes of deep neural networks (DNNs) to edge devices, such as federated learning [33] and model compression [13][35].”
BackgroundOverparametrization of HyperNetworks at Fixed FLOP-Count Enables Fast Neural Image Enhancement
Müller, L. K. (2021) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw53098.2021.00038 · Cited by 8
“An excellent overview is given in [7].”
Compute-in-RRAM with Limited On-chip Resources
Lu, A., Peng, X., & Yu, S. (2021) · IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS) · doi:10.1109/aicas51828.2021.9458481 · Cited by 6
“State-of-the-art deep learning algorithms have demonstrated remarkable effectiveness for various artificial intelligence applications such as image processing, speech recognition and language translation, etc. [1].”
BackgroundRealistic acceleration of neural networks with fine-grained tensor decomposition
Lv, R., Wang, D., Zheng, J., Xie, Y., & Yang, Z.-X. (2022) · Neurocomputing · doi:10.1016/j.neucom.2022.09.057 · Cited by 6
Self-citationTowards human-leveled vision systems
Ding, J., & Huang, T. (2024) · Science China Technological Sciences · doi:10.1007/s11431-024-2762-5 · Cited by 5
“The tendency towards deeper and more complex neural networks has prompted the rise of compression technology [57].”
Distributed Inference Models and Algorithms for Heterogeneous Edge Systems Using Deep Learning
Yuan, Q., & Li, Z. (2025) · Applied Sciences · doi:10.3390/app15031097 · Cited by 7
Edge AI in Vehicle Modules Using Heterogeneous Networks: Research Trends and Future Directions
Ryu, D., Kang, Y., Jeong, M., Batzorig, M., & Yim, K. (2025) · Lecture notes on data engineering and communications technologies · doi:10.1007/978-3-031-96093-2_31 · Cited by 6
Liu, Y., & Shi, L. (2025) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2025.3609087 · Cited by 5
Self-citationA robust stacked neural network approach for early and accurate breast cancer diagnosis
Li, X., Gao, M., Zhang, C., Ma, G., Zhang, Q., Meng, W., Yuan, T., Wang, Y., & Li, Z. (2025) · Frontiers in Medicine · doi:10.3389/fmed.2025.1644857 · Cited by 7
“First, model pruning and quantization techniques can reduce the number of model parameters and storage requirements, thereby significantly shortening inference latency while maintaining performance close to the original model (50).”
Benchmarking deep learning models for ET0 forecasting on edge devices
Luo, K., Lim, C. S., Ishak, M. H. I., Mahmud, M. S. A., & Ba, N. (2025) · Hydrology research · doi:10.2166/nh.2025.130 · Cited by 5
Ali, M. B. G., & Smiti, A. (2026) · Multimedia Tools and Applications · doi:10.1007/s11042-026-21266-x · Cited by 5
Chou, H.-H., Chiu, C., & Liao, Y.-P. (2021) · APSIPA Transactions on Signal and Information Processing · doi:10.1017/atsip.2021.16 · Cited by 6
A Lightweight Conditional Convolutional Neural Network for Hyperspectral Image Classification
Wu, L., Wang, H., & Wang, H. (2023) · Photogrammetric Engineering & Remote Sensing · doi:10.14358/pers.22-00130r2 · Cited by 5
Toward Inference Delivery Networks: Distributing Machine Learning With Optimality Guarantees
Salem, T. S., Castellano, G., Neglia, G., Pianese, F., & Araldo, A. (2023) · IEEE/ACM Transactions on Networking · doi:10.1109/tnet.2023.3305922 · Cited by 10
“Therefore, we study the novel problem of how to deploy the available ML models on the available IDN nodes, where a deployment strategy consists in two coupled decisions: (i) where to place models for serving a certain task and (ii) how to select their size/complexity among the available…”
MethodEfficient Federated Learning using Random Pruning in Resource-Constrained Edge Intelligence Networks
Chen, C., Jiang, B., Liu, S., Li, C., Wu, C., & Yin, R. (2023) · Globecom · doi:10.1109/globecom54140.2023.10437051 · Cited by 5
“To address this issue, model compression has been widely considered to accelerate the training process recently [1], [2].”
BackgroundImage inpainting based on tensor ring decomposition with generative adversarial network
Yuan, J., Wu, H., Zhao, L., & Wu, F. (2024) · Signal Image and Video Processing · doi:10.1007/s11760-024-03415-7 · Cited by 5
“Compared with the methods mentioned above, the tensor decomposition method [21], which has a solid theoretical foundation, has garnered increasing attention in recent years due to its ease of implementation and high compression ratio [40].”
Haiyan, K., & Jiakang, W. (2024) · Complex & Intelligent Systems · doi:10.1007/s40747-024-01573-2 · Cited by 5
“One approach is to implement compact yet precise models, including streamlined architecture designs, model compression [11], and pruning [12], directly reducing the size of model updates.”
CROSS: Compiler-Driven Optimization of Sparse DNNs Using Sparse/Dense Computation Kernels
Liu, F., Huang, S., Yang, N., Wang, Z., Li, H., & Jiang, L. (2025) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca61900.2025.00076 · Cited by 11
“Notably, this level of sparsity can be achieved by almost all DNNs with negligible impact on accuracy [9], [55].”
Theoretical guarantees for low-rank compression of deep neural networks
Zhang, S., & Saab, R. (2025) · Applied and Computational Harmonic Analysis · doi:10.1016/j.acha.2025.101837 · Cited by 6
“Consequently, the demand for model compression techniques has grown, particularly in contexts where storage efficiency and adaptability to mobile devices are crucial [9, 54].”
An impact of tensor-based data compression methods on deep neural network accuracy
Grabek, J., & Cyganek, B. (2021) · Annals of Computer Science and Information Systems · doi:10.15439/2021f127 · Cited by 5
“Tensors extend the notion of vectors and matrices into higher-dimensional objects [2], [3], [21]–[23].”
BackgroundMethodAntiDoteX: Attention-Based Dynamic Optimization for Neural Network Runtime Efficiency
Yu, F., Xu, Z., Liu, C., Stamoulis, D. S., Wang, D., Wang, Y., & Chen, X. (2022) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3144616 · Cited by 6
“However, the increasingly larger model structure also introduces tremendous computation load, causing considerable performance issues [8]–[11].”
BackgroundAdaptive federated learning algorithm based on evolution strategies
公, 茂., 高, 原., 王, 炯., 张, 元., 王, 善., & 谢, 飞. (2022) · Scientia Sinica Informationis · doi:10.1360/ssi-2021-0190 · Cited by 5
Fast visual tracking with lightweight Siamese network and template-guided learning
Zhang, Y., Liu, G., Huang, H., & Xiong, R. (2022) · Knowledge-Based Systems · doi:10.1016/j.knosys.2022.110037 · Cited by 5
Xiang, W., & Shao, Z. (2022) · IEEE Conference on Decision and Control (CDC) · doi:10.1109/cdc51059.2022.9992984 · Cited by 5
“More results on neural network model reduction can be found in a recent survey [8].”
BackgroundFIANCEE: Faster Inference of Adversarial Networks via Conditional Early Exits
Karpikova, P., Radionova, E., Yaschenko, A., Andrei, S., Kostyushko, L., Fabbricatore, R., & Aleksei, I. (2023) · 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/cvpr52729.2023.01158 · Cited by 6
“2 [14] Lei Deng, Guoqi Li, Song Han, Luping Shi, and Yuan Xie.”
BackgroundLlanza, A., Keddous, F. E., Shvai, N., & Nakib, A. (2023) · Conference proceedings/Conference proceedings - IEEE International Conference on Systems, Man, and Cybernetics · doi:10.1109/smc53992.2023.10394477 · Cited by 5
Liu, W., Zhang, M., Shi, C., Zhang, N., & Liu, J. (2024) · Neural Processing Letters · doi:10.1007/s11063-024-11465-8 · Cited by 5
“DNNs compressed methods are divided into five categories according to different compression mechanisms [9]: pruning [10, 11], quantization [12, 13], knowledge distillation [14, 15], low-rank decomposition [16, 17], and network architecture search [18–20].”
Reweighted Alternating Direction Method of Multipliers for DNN weight pruning
Yuan, M., Du, L., Jiang, F., Bai, J., & Chen, G. (2024) · Neural Networks · doi:10.1016/j.neunet.2024.106534 · Cited by 6
Teo, N. Q. X., Hall-Chen, V. H., Barada, K. K., Ng, R. J. H., Gu, L., Yeoh, A. K., Pratt, Q., Garbet, X., & Rhodes, T. L. (2024) · Review of Scientific Instruments · doi:10.1063/5.0215748 · Cited by 5
“There are various techniques to achieve this, such as sparsification 28 to reduce the model size.”
MethodShrinking the giants: Paving the way for TinyAI
Nakhle, F. (2024) · Device · doi:10.1016/j.device.2024.100411 · Cited by 5
Rapid Deployment of DNNs for Edge Computing via Structured Pruning at Initialization
Eccles, B. J., Wong, L., & Varghese, B. (2024) · IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid) · doi:10.1109/ccgrid59990.2024.00044 · Cited by 7
“Therefore, model compression methods that reduce the resource requirements of training and inference while preserving task accuracy are used [7].”
MethodWinata, A., Andryani, N. A. C., Gunawan, A. A. S., & Gaol, F. L. (2025) · IEEE Access · doi:10.1109/access.2025.3581225 · Cited by 5
“While using an efficient model has a simple practical approach, there is a tradeoff between model size and model performance [6].”
Towards to an Embedded Edge AI Implementation for Longitudinal Rip Detection in Conveyor Belt
Klippel, E., Oliveira, R. A., Maslov, D., Bianchi, A. G. C., Delabrida, S. E., & Garrocho, C. T. B. (2020) · doi:10.5753/sbesc_estendido.2020.13096 · Cited by 5
“The optimization of its operation is achieved through specific processing and memory architectures for operation with deep neural networks [9].”
BackgroundMixture of Deterministic and Stochastic Quantization Schemes for Lightweight CNN
Kim, S.-R., & Kim, H. (2020) · International SoC Design Conference · doi:10.1109/isocc50952.2020.9332958 · Cited by 5
“In this paper, the two methods of data quantization, deterministic and stochastic quantization [9], are appropriately mixed based on the impact on the accuracy drop, which achieves considerable accuracy enhancement compared to each conventional method.”
MethodA New Clustering-Based Technique for the Acceleration of Deep Convolutional Networks
Pikoulis, E. V., Mavrokefalidis, C., & Lalos, A. S. (2020) · International Conference on Machine Learning and Applications · doi:10.1109/icmla51294.2020.00222 · Cited by 10
Contactless Human Monitoring: Challenges and Future Direction
Mahbub, U., Rahman, T., & Ahad, M. A. R. (2021) · Intelligent systems reference library · doi:10.1007/978-3-030-68590-4_12 · Cited by 6
Accelerating 3D scene analysis for autonomous driving on embedded AI computing platforms
Nousias, S., Pikoulis, E., Mavrokefalidis, C., Lalos, A. S., & Μουστάκας, Κ. (2021) · IFIP/IEEE International Conference on Very Large Scale Integration · doi:10.1109/vlsi-soc53125.2021.9606990 · Cited by 5
“The MCA literature is thoroughly described in several recent survey papers like [13], [14] and [15], where the presented techniques can be utilized on the algorithmic or the hardware level, while, currently, there are also methods that take into account both algorithmic and hardware aspects.”
Efficient neural network representations for energy data analytics on embedded systems
Bouchur, M., & Reinhardt, A. (2022) · ACM International Conference on Future Energy Systems (e-Energy) · doi:10.1145/3538637.3538842 · Cited by 5
“Consequently, several compression techniques have emerged in recent years to achieve a trade-off between processing efficiency, memory demand, and achievable accuracy [14].”
BackgroundRedundancy Pruning for Binary Hyperdimensional Computing Architectures
Antonio, R. A. G., & Alvarez, A. B. (2022) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas48785.2022.9937640 · Cited by 5
“State-of-the-art machine learning (ML) and deep neural network (DNN) algorithms require tons of computing power to achieve top-notch accuracy performance [1].”
BackgroundQuPeL: Quantized Personalization with Applications to Federated LearningPreprint
Ozkara, K., Singh, N., Data, D., & Diggavi, S. N. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.11786 · Cited by 5
“Recent work adopted different approaches for learning personalized models: (i) Combine global and local models throughout the training Deng et al. (2020); Hanzely & Richtárik (2020); Mansour et al. (2020); (ii) first learn a global model and then personalize it locally by updating it using clients’ local data Fallah e…”
BackgroundMethodInfluentialDynamical Isometry: The Missing Ingredient for Neural Network PruningPreprint
Wang, H., Qin, C., Bai, Y., & Fu, Y. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2105.05916 · Cited by 5
“For more comprehensive literature, we refer interested readers to several surveys: an outdated one [39], some recent surveys of pruning alone [12, 1] or pruning as a sub-topic under the general umbrella of model compression and acceleration [46, 3, 4, 7].”
BackgroundFinding Everything within Random Binary NetworksPreprint
Sreenivasan, K., Rajput, S., Sohn, J.-Y., & Papailiopoulos, D. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.08996 · Cited by 6
“A rich and long body of research, dating back to the 80s, shows that one can prune most networks to a tiny fraction of their size, while maintaining high accuracy (Mozer and Smolensky, 1989; Hassibi and Stork, 1993; Levin et al., 1994; LeCun et al., 1990; Han et al., 2015b,a; Li et al., 2016; Wen et al., 2016; Hubara…”
BackgroundA Review of AIoT-based Edge Devices and Lightweight DeploymentPreprint
Sun, K., Wang, X., & Zhao, Q. (2023) · doi:10.36227/techrxiv.21687248.v2 · Cited by 5
Model Quantization and Hardware Acceleration for Vision Transformers: A Comprehensive Survey
Du, D., Gong, G., & Chu, X. (2024) · arXiv · doi:10.48550/arxiv.2405.00314 · Cited by 4
“For a deeper understanding of the foundational architecture of hardware accelerators, we direct readers to [75], [76], [77].”
BackgroundASVD: Activation-aware Singular Value Decomposition for Compressing Large Language Models
Yuan, Z., Shang, Y., Song, Y., Yang, D., Wu, Q., Yan, Y., & Sun, G. (2023) · arXiv · doi:10.48550/arxiv.2312.05821 · Cited by 4
“Furthermore, low-rank decomposition is complementary to existing LLM compression techniques by further compressing quantized or pruned models, enhancing overall efficiency [Cheng et al., 2017, Deng et al., 2020].”
MethodEfficient and Compact Representations of Deep Neural Networks via Entropy Coding
Marinò, G. C., Furia, F., Malchiodi, D., & Frasca, M. (2023) · IEEE Access · doi:10.1109/access.2023.3317293 · Cited by 4
“Many studies have already shown how most state-of-the-art DNNs in various contexts can often undergo to lossy compression to the detriment of negligible accuracy drops [10], [11], [12], which encouraged researchers to explore the field of DNNs compression.”
BackgroundCompressed Test Pattern Generation for Deep Neural Networks
Moussa, D. A., Hefenbrock, M., & Tahoori, M. B. (2024) · IEEE Transactions on Computers · doi:10.1109/tc.2024.3457738 · Cited by 6
“While many related works have investigated test generation and compaction for hardware testing and their associated challenges [12], [13], [14], [15], [16], other studies employed software-based solutions to enhance flexibility and efficiency in functional testing.”
Hourglass: Enabling Efficient Split Federated Learning with Data Parallelism
He, Q., Wang, K., Dong, Z., Yuan, L., Chen, F., Jin, H., & Yang, Y. J. (2025) · European Conference on Computer Systems · doi:10.1145/3689031.3717467 · Cited by 9
“However, model compression often results in a substantial loss in model accuracy [13, 22, 29, 74].”
TDJSCC: Low Complexity and Bandwidth Efficient Deep Joint Source-Channel Coding With OFDM
Xu, M., Lam, C., Liang, Y., Ng, B. K., Yuan, X., & Im, S. (2025) · IEEE Access · doi:10.1109/access.2025.3602059 · Cited by 4
“Firstly, to find the first order tensor Y 1 , we assign the value of K to Y (1) and then perform the tensorizing SVD algorithm as follows : where The symbol ∗ denotes the tensor product operation [33].”
ICE-Pruning: An Iterative Cost-Efficient Pruning Pipeline for Deep Neural Networks
Hu, W., Henderson, P., & Cano, J. (2025) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn64981.2025.11227410 · Cited by 6
“With the increasing attention on pruning research [10], several approaches have been developed [9], [11].”
Performance analysis of Convolutional Neural Networks on edge devices for Computer Vision tasks
Bricola, A., Noceti, N., & D’Agostino, D. (2026) · Array · doi:10.1016/j.array.2026.100692 · Cited by 5
Vo, D. T. D., Nguyen, V. D., Nguyen, T. H. M., & Võ, M. H. (2026) · IEEE Access · doi:10.1109/access.2026.3681176 · Cited by 4
“To address this hardware barrier [15], Knowledge Distillation (KD) [16] has established itself as a critical compression paradigm.”
Target-Aware Feature Bottleneck for Real-Time Visual Tracking
Choi, J. (2023) · Applied Sciences · doi:10.3390/app131810198 · Cited by 4
FlexBCM: Hybrid Block-Circulant Neural Network and Accelerator Co-Search on FPGAs
Lou, W., Qin, Y., Wang, X., Gong, L., Wang, C., & Zhou, X. (2024) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2024.3439488 · Cited by 4
“To this end, model compression has emerged as an effective method to reduce model redundancy [4], [5], [6].”
MethodGarg, S., & Singh, P. (2025) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2025.3527402 · Cited by 4
“In this direction, several approaches have been used that includes pruning, quantization, binarization and the most promising one, model distillation [3], [4], [5], [6], [7].”
Exploring advanced deep learning approaches in cardiac image analysis: A comprehensive review
Boukhamla, A., Azizi, N., Belhouari, S. B., & Dey, N. (2025) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2025.110708 · Cited by 5
Mhaouch, A., Gtifa, W., Althobaiti, T., Faraj, H., & Machhout, M. (2025) · Computers, materials & continua/Computers, materials & continua (Print) · doi:10.32604/cmc.2025.065525 · Cited by 4
Layer Redundancy Aware DNN Model Repository Planning for Fast Model Download in Edge Cloud
Geng, H., Li, Y., Wang, S., Gu, L., & Zeng, D. (2025) · IEEE Transactions on Cloud Computing · doi:10.1109/tcc.2025.3591482 · Cited by 4
“To diminish the downloading volume, model compression techniques [25], including model pruning [26], [27], [28] by removing non-critical mode structures and model quantization methods [29], [30] by translating the model parameters from higher to lower precision formats.”
Li, J., & Wu, Q. (2025) · Cluster Computing · doi:10.1007/s10586-024-05032-y · Cited by 6
Halftoning with Multi-Agent Deep Reinforcement Learning
Jiang, H., Xiong, D., Jiang, X., Yin, A., Ding, L., & Huang, K. (2022) · Proceedings - International Conference on Image Processing · doi:10.1109/icip46576.2022.9897198 · Cited by 4
“It is possible to close the speed gap between our model and error diffusion approaches by designing a more efficient CNN or using compression techniques [25], and we leave it for future work.”
MethodAccordion: A Communication-Aware Machine Learning Framework for Next Generation Networks
Ayed, F., Domenico, A. D., García‐Rodríguez, A., & López‐Pérez, D. (2023) · IEEE Communications Magazine · doi:10.1109/mcom.001.2200358 · Cited by 7
“The main techniques used to reduce the communication requirements and the computational complexity of AI/ML models are [10], [11]:”
MethodCompressing Deep Neural Networks Using Explainable AI
Soroush, K., Raji, M., & Ghavami, B. (2023) · International Conference on Computer and Knowledge Engineering · doi:10.1109/iccke60553.2023.10326237 · Cited by 8
“However, traditional compression methods [10] finding less important weights for pruning or quantizing, leading to inefficient compressed DNN models.”
MethodZhu, J., Wang, L., Han, X. G., Liu, A., & Xie, T. (2024) · IEEE Transactions on Software Engineering · doi:10.1109/tse.2023.3348515 · Cited by 11
“[13], [14]; compared to big models (often stored in a well-maintained server), an adversary 2 thereby has much more opportunities to gain access to compressed models and attack them.”
BackgroundA lightweight real-time detection method of small objects for home service robots
Zhang, T., Zhao, F., Zou, Y., & Zheng, J. (2024) · Machine Vision and Applications · doi:10.1007/s00138-024-01611-6 · Cited by 4
“Network structure design is an optimization method that operates independently of a pre-trained network, and is widely used because of its excellent lightweight effect and high adaptability [36].”
Ghavami, B., Sajadi, M., Shannon, L., & Wilton, S. J. E. (2024) · IEEE Computer Society Annual Symposium on VLSI · doi:10.1109/isvlsi61997.2024.00049 · Cited by 5
“However, in modern DNN models, a common practice involves reducing data precision, a technique known as quantization , which significantly lowers computational and memory requirements [13].”
MethodLearnable Heterogeneous Convolution: Learning both topology and strength
Zhao, R., Wu, Z., & Zhang, Q. (2021) · Neural Networks · doi:10.1016/j.neunet.2021.03.038 · Cited by 7
Towards privacy aware deep learning for embedded systems
Duddu, V., Boutet, A., & Shejwalkar, V. (2022) · ACM Symposium on Applied Computing · doi:10.1145/3477314.3507128 · Cited by 6
“for low precision NNs [37] while skipping computations to improve performance for compressed models [6].”
BackgroundNeural Networks Reduction via Lumping
Ressi, D., Romanello, R., Piazza, C., & Rossi, S. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-27181-6_6 · Cited by 4
“As these models have evolved to process large amount of data or to solve complicated tasks, their complexity has also increased at same pace [10].”
Optimization Ways in Neural Network Compression
Zhou, R., & Quan, P. (2023) · Procedia Computer Science · doi:10.1016/j.procs.2023.08.125 · Cited by 4
Multi-view learning-based heterogeneous network representation learning
Chen, L., Li, Y., & Deng, X. (2023) · Journal of King Saud University - Computer and Information Sciences · doi:10.1016/j.jksuci.2023.101855 · Cited by 4
“These features are coupled together in a disordered manner that is difficult to extract accurately (Deng et al., 2020; Li et al., 2022).”
Accelerating Deep Neural Networks via Semi-Structured Activation Sparsity
Grimaldi, M., Ganji, D. C., Lazarevich, I. A., & Sah, S. (2023) · IEEE International Conference on Computer Vision Workshops · doi:10.1109/iccvw60793.2023.00127 · Cited by 17
“Compression and acceleration of such models is an active research area aimed at bridging this gap [3] and could be generally categorized into pruning [29,34,52], tensor decomposition [33], quantization [7,41], development of lightweight neural networks [21,22,37], and runtime optimizations [2,15].”
BackgroundLiu, G., Bilal, M., Xu, X., & Xia, X. (2024) · IEEE Communications Magazine · doi:10.1109/mcom.001.2300697 · Cited by 4
“This enhancement enables the model to conduct inference more efficiently [12].”
Towards Memory-Efficient Neural Networks via Multi-Level in situ Generation
Gu, J., Zhu, H., Feng, C., Liu, M., Jiang, Z., Chen, R. T., & Pan, D. Z. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00518 · Cited by 4
“Previous efforts towards memory-efficient accelerator designs focus on weight quantization [20, 38, 60], pruning with sparsity exploration [9, 20, 21, 30, 50, 55], structured weight matrices [11,14,31,49], slim architectures [2,7,25], better hardware scheduling [34, 56], low-rank approximation [10,…”
BackgroundGECKO: Reconciling Privacy, Accuracy and Efficiency in Embedded Deep LearningPreprint
Duddu, V., Boutet, A., & Shejwalkar, V. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.00912 · Cited by 4
“for low precision NNs [37] while skipping computations to improve performance for compressed models [6].”
BackgroundCompact representations of convolutional neural networks via weight pruning and quantizationPreprint
Marinò, G. C., Petrini, A., Malchiodi, D., & Frasca, M. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2108.12704 · Cited by 4
“interested to a thorough review of compression methods for NNs can refer for instance to [6], [7].”
BackgroundAskewSGD : An Annealed interval-constrained Optimisation method to train Quantized Neural NetworksPreprint
Leconte, L., Schechtman, S., & Moulines, É. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2211.03741 · Cited by 4
A Review of AIoT-based Edge Devices and Lightweight DeploymentPreprint
Sun, K., Wang, X., & Zhao, Q. (2022) · doi:10.36227/techrxiv.21687248 · Cited by 4
Fida, K., Abbasi, U., Adnan, M., Iqbal, M. S., Zahid, H., & Mohamed, S. E. G. (2024) · Engineering Archive (engrXiv) · doi:10.31224/4069 · Cited by 4
Ensuring Bidirectional Privacy on Wireless Split Inference Systems
Sa, C.-C., Cheng, L.-C., Chung, H.-H., Chiu, T.-C., Wang, C.-Y., Pang, A., & Chen, S.-T. (2024) · IEEE Wireless Communications · doi:10.1109/mwc.014.2300400 · Cited by 3
“In a system where raw output is isolated from attackers, label inference attacks can occur as long as there is other side information, such as the gradients in the training process”
BackgroundApproximation-Aware Training for Efficient Neural Network Inference on MRAM Based CiM Architecture
Nehete, H., Soni, S., Reddy, T. K., Raman, B., & Kaushik, B. K. (2024) · IEEE Open Journal of Nanotechnology · doi:10.1109/ojnano.2024.3524265 · Cited by 3
An efficient quantized GEMV implementation for large language models inference with matrix core
Zhang, Y., Lu, L., Zhao, R., Guo, Y., & Yang, Z. (2025) · The Journal of Supercomputing · doi:10.1007/s11227-025-06993-6 · Cited by 4
“To address such a concern, a series of model compression methods, such as quantization [8, 9], low-rank factorization [10], pruning, and knowledge distillation [11], have been proposed to reduce the pressure on the memory foot-print [12, 13].”
Training 4.6-Bit Convolutional Neural Networks with a HardTanh Activation Function
Trusov, A. V. (2025) · Pattern Recognition and Image Analysis · doi:10.1134/s105466182470144x · Cited by 3
“…one implies increasing the scale and complexity of models running on dedicated servers and cloud platforms [4], while the second one is about simplifying and reducing neural networks for them to be integrated into systems running on users’ devices such as cell phones and other handheld devices [9].”
Adaptive model switching of collaborative inference for multi-CNN streams in UAV swarm
Li, Y., Qu, Y., Dong, C., Qin, Z., Zhang, L., & Wu, Q. (2025) · Chinese Journal of Aeronautics · doi:10.1016/j.cja.2025.103564 · Cited by 5
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.