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 401–500 of 1,002
Stochastic batch size for adaptive regularization in deep network optimization
Nakamura, K., Soatto, S., & Hong, B. (2022) · Pattern Recognition · doi:10.1016/j.patcog.2022.108776 · Cited by 10
VLCQ: Post-training quantization for deep neural networks using variable length coding
Abdel‐Salam, R., Abdel-Gawad, A. H., & Wassal, A. G. (2024) · Future Generation Computer Systems · doi:10.1016/j.future.2024.107654 · Cited by 10
Heterogeneous Binary Pixel Difference Networks for Remote Sensing Object Detection
Zhan, J., Bai, L., Zhang, J., Liu, T., Shi, F., Liu, Y., & Liu, L. (2024) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2024.3520161 · Cited by 10
Lightweight CSI-Based Human Activity Recognition for Multitask IoT Applications
Liu, C., Miao, F., Chen, Y., Wu, J., Wan, H., Tang, T., Ohtsuki, T. O., Gui, G., & Sari, H. (2025) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3580141 · Cited by 9
“These indicators directly reflect the model’s suitability for real-time deployment on edge or IoT devices [33].”
Deep Learning-Based Speech Enhancement for Robust Sound Classification in Security Systems
Mensah, S. Y., Zhang, T., Mahmud, N. A., & Geng, Y. (2025) · Electronics · doi:10.3390/electronics14132643 · Cited by 10
EIoT: Embodied Intelligence of Things
Liu, Y., Wang, X., Liu, Y., Liu, K., Tong, S., Yuan, J., & Li, L. (2025) · Tsinghua Science & Technology · doi:10.26599/tst.2025.9010127 · Cited by 9
Sahu, I., Ukil, A., Khandelwal, S., & Pal, A. (2022) · Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) · doi:10.1109/embc48229.2022.9871259 · Cited by 9
“It achieves around 142x model size reduction without classification penalty, whereas best of the class pruning method (fine-tuned global pruning [10, 11, 14]) cannot reduce more than 48x times.”
MethodExploring Winograd Convolution for Cost-Effective Neural Network Fault Tolerance
Xue, X., Liu, C., Лю, Б., Huang, H., Wang, Y., Luo, T., Zhang, L., Li, H., & Li, X. (2023) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2023.3306894 · Cited by 11
Slim-Tasnet: A Slimmable Neural Network for Speech Separation
Elminshawi, M., Chetupalli, S. R., & Habets, E. A. P. (2023) · IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA) · doi:10.1109/waspaa58266.2023.10248143 · Cited by 9
Jin, R., Yuan, W., Zhang, J., Shi, J., & Chi, N. (2024) · Optics Express · doi:10.1364/oe.510449 · Cited by 10
Maximum Output Discrepancy Computation for Convolutional Neural Network Compression
Mo, Z., & Xiang, W. (2023) · Information Sciences · doi:10.1016/j.ins.2024.120367 · Cited by 9
Wang, D., Lu, Y., Yang, X., Liu, D., Yang, X., & Yang, J. (2024) · Applied Soft Computing · doi:10.1016/j.asoc.2024.112640 · Cited by 9
McEnroe, P., Wang, S., & Liyanage, M. (2024) · IEEE International Conference on Intelligent Transportation Systems (ITSC) · doi:10.1109/itsc58415.2024.10919742 · Cited by 9
“…entire layers from a network, neuron pruning approaches remove entire neurons from a network, channel pruning approaches remove entire channels (i.e. feature maps) from convolutional layers and filter pruning approaches remove entire filters (i.e. kernels) from convolutional layers [16], [17].”
A Survey on Foundation Models for Personalized Federated Intelligence
Qiao, Y., Le, H. Q., Raha, A. D., Tran, P.-N., Adhikary, A., Zhang, M., Nguyen, L. X., Huh, E., Niyato, D. T., & Hong, C. S. (2026) · ACM Computing Surveys · doi:10.1145/3844947 · Cited by 11
A Construction Kit for Efficient Low Power Neural Network Accelerator Designs
Jokic, P., Azarkhish, E., Bonetti, A., Pons, M., Emery, S., & Benini, L. (2022) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3520127 · Cited by 11
“A survey on hardware acceleration of compressed models can be found in [155].”
BackgroundYu, T., Wu, B., Chen, K., Yan, C. C., & Liu, W. (2022) · ACM Great Lakes Symposium on VLSI (GLSVLSI) · doi:10.1145/3526241.3530318 · Cited by 9
“Especially in edge-end applications, relying simply on existing unstructured pruning schemes not only fails to meet the demand for smaller model volumes, but also fails to guarantee processing accuracy and hardware efficiency [8].”
BackgroundLRP-based Policy Pruning and Distillation of Reinforcement Learning Agents for Embedded Systems
Xu, R., Luan, S., Gu, Z., Zhao, Q., & Chen, G. (2022) · IEEE International Symposium on Real-Time Distributed Computing (ISORC) · doi:10.1109/isorc52572.2022.9812837 · Cited by 13
“Unstructured pruning techniques remove individual weights and can achieve high degrees of sparsity, but the resulting unstructured sparse models are not hardware-friendly, and cannot achieve significant speedup in inference time on typical hardware accelerators [6]–[8] for deep learning.”
BackgroundFilippini, F., Lattuada, M., Ciavotta, M., Jahani, A., Ardagna, D., & Amaldi, E. (2022) · IEEE Transactions on Services Computing · doi:10.1109/tsc.2022.3188440 · Cited by 9
“pression and acceleration techniques [1] are employed to reduce the dimensionality of the training problems.”
MethodNeuron Specific Pruning for Communication Efficient Federated Learning
Kumar, G., & Toshniwal, D. (2022) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/3511808.3557658 · Cited by 9
“Recent work on DNNmodel compression in centralized model training has achieved an incredible compression rate [3].”
BackgroundLuan, S., Gu, Z., Xu, R., Zhao, Q., & Chen, G. (2022) · Concurrency and Computation Practice and Experience · doi:10.1002/cpe.7351 · Cited by 10
“The suffix robust denotes that the agent is trained with the robust training algorithm RADIAL-RL with the loss function defined in Equation (11), and the results are drawn with solid lines; the suffix std denotes that the agent is trained with the standard DQN, and the results are drawn with dotted lines.”
MethodInfluentialNeural-Network-Assisted Packet Accelerators for Internet of Things Network Systems
Nwadiugwu, W., Ejaz, W., Kaneko, M., & Anpalagan, A. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3265344 · Cited by 9
“Recent ML-based data compression studies [36], [37], [38] have gained traction.”
BackgroundMetaML: Automating Customizable Cross-Stage Design-Flow for Deep Learning Acceleration
Que, Z., Liu, S., Rognlien, M., Guo, C., Coutinho, J. G. F., & Luk, W. W. (2023) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl60245.2023.00042 · Cited by 10
“Achieving optimal results for a given problem is complex [9].”
BackgroundEfficient object tracking algorithm based on lightweight Siamese networks
Feng, Z., & Wang, H. (2024) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2024.107976 · Cited by 9
Joint-Way Compression for LDPC Neural Decoding Algorithm With Tensor-Ring Decomposition
Liang, Y., Lam, C.-T., & Ng, B. (2023) · IEEE Access · doi:10.1109/access.2023.3252907 · Cited by 10
“This low-complexity technique using multiple compression methods is called joint-way compression [12].”
MethodPost-Training Quantization for Energy Efficient Realization of Deep Neural Networks
Latotzke, C., Balim, B., & Gemmeke, T. (2022) · 2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA) · doi:10.1109/icmla55696.2022.00243 · Cited by 8
“There exist a variety of compression techniques for DNNs to increase efficiency during inference [18], [19].”
BackgroundFPGA-Based Spiking Neural Networks
Mehrabi, A., & André, V. S. (2024) · Artificial intelligence · doi:10.5772/intechopen.1006168 · Cited by 8
Precision decarbonization for clean dairy farming with digital twin and edge intelligence
Feng, C., Reed, K. F., Giordano, J. O., & You, F. (2025) · Nexus · doi:10.1016/j.ynexs.2025.100105 · Cited by 8
On-Device Deep Multi-Task Inference via Multi-Task Zipping
He, X., Wang, X., Zhou, Z., Wu, J., Yang, Z., & Thiele, L. (2021) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2021.3124306 · Cited by 11
“Neuron-level pruning is more desirable since it leads to regular sparsity in the pruned networks, and thus avoids the need for customised hardware [11].”
BackgroundMethodA Technique for Approximate Communication in Network-on-Chips for Image Classification
Chen, Y., Liu, S., Lombardi, F., & Louri, A. (2022) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2022.3162165 · Cited by 8
Li, T., Ma, Y., & Endoh, T. (2023) · Neural Computing and Applications · doi:10.1007/s00521-023-08280-y · Cited by 8
“The weight pruning aims to remove the redundant weights while feature map pruning decreases both feature maps and weights [30].”
BackgroundLightweight TCN-Based Spatial-Temporal Channel Extrapolation for RIS-Aided Communication
Gong, T., Zhang, S., Xu, M., Li, Z., & Dobre, O. A. (2024) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2024.3381217 · Cited by 8
“Besides, in the resource-constrained application, the DL algorithm is limited due to its great demand for computing resources [12].”
BackgroundEvolving filter criteria for randomly initialized network pruning in image classification
Chen, X., Liu, C., Peng, H., Lin, J., Gong, Y., Chen, Y., Peng, D., & Geng, X. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.127872 · Cited by 10
A&B BNN: Add&Bit-Operation-Only Hardware-Friendly Binary Neural Network
Ma, R., Qiao, G., Liu, Y., Meng, L., Ning, N., Liu, Y., & Hu, S. (2024) · 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/cvpr52733.2024.00545 · Cited by 8
“However, as the scale of deep neural networks (DNNs) expands, the substantial computational and storage requirements make them feasible only for running on powerful but expensive GPUs, leaving edge devices unattainable [11, 12].”
BackgroundZhao, B., Qin, Z., Wu, Y., Song, Y., Yu, H., & Gao, L. (2024) · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · doi:10.1109/jstars.2024.3483749 · Cited by 8
“Moststudieshavefocusedonreducing thenumberofparametersandcomputations[2].”
One-core neuron deep learning for time series prediction
Peng, H., Chen, P., Yang, N., Aihara, K., Liu, R., & Chen, L. (2024) · National Science Review · doi:10.1093/nsr/nwae441 · Cited by 13
Combining Stochastic Defenses to Resist Gradient Inversion: An Ablation Study
Scheliga, D., Mäder, P., & Seeland, M. (2025) · Applied Artificial Intelligence · doi:10.1080/08839514.2025.2565165 · Cited by 8
“This is typically achieved through GP (Deng et al. 2020; Lin et al. 2017; Tsuzuku, Imachi, and Akiba 2018), which removes less informative elements from the training gradients before transmission to the server.”
Dictionary Pair-based Data-Free Fast Deep Neural Network Compression
Gao, Y., Zhang, Z., Zhang, H., Zhao, M., Yang, Y., & Wang, M. (2021) · Proceedings · doi:10.1109/icdm51629.2021.00022 · Cited by 9
Structured Dynamic Precision for Deep Neural Networks Quantization
Huang, K., Li, B., Xiong, D., Jiang, H., Jiang, X., Yan, X., Claesen, L., Liu, D., Chen, J., & Liu, Z. (2022) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3549535 · Cited by 10
“The number of bits of HAWQ is limited to a range of [4, 8].”
BackgroundMethodJoint Architecture Design and Workload Partitioning for DNN Inference on Industrial IoT Clusters
Fang, W. W., Xu, W., Yu, C., & Xiong, N. N. (2022) · ACM Transactions on Internet Technology · doi:10.1145/3551638 · Cited by 11
“A number of techniques have been proposed to address how to accelerate DNN model inference on resourceconstrained edge devices [1, 35].”
BackgroundMethodInfluentialXiang, W., & Shao, Z. (2022) · Proceedings of the ... American Control Conference/Proceedings of the American Control Conference · doi:10.23919/acc53348.2022.9867845 · Cited by 8
“More neural network compression results can be found in the recent survey [9] and references therein.”
MethodResultJin, X., Ba, W., Wang, L., Zhang, T., Zhang, X., Li, S., Rao, Y., & Liu, L. (2022) · ACS Omega · doi:10.1021/acsomega.2c03596 · Cited by 8
Song, Y., Sha, E. H. ‐., Zhuge, Q., Xu, R., Xu, X., Li, B., & Yang, L. (2022) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2022.102810 · Cited by 8
Lin, Y., Niu, L., Yang, X. H., & Zhou, R. (2023) · Pattern Recognition · doi:10.1016/j.patcog.2023.109556 · Cited by 9
Rutishauser, G., Scherer, M., Fischer, T., & Benini, L. (2023) · Future Generation Computer Systems · doi:10.1016/j.future.2023.07.017 · Cited by 11
OCAP: On-device Class-Aware Pruning for personalized edge DNN models
Ma, Y.-D., Zhi-chao, Z., Liu, D., He, Z., & Zhou, W. (2023) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2023.102956 · Cited by 8
Sagitta: An Energy-Efficient Sparse 3D-CNN Accelerator for Real-Time 3-D Understanding
Zhou, C., Liu, M., Qiu, S., Cao, X., Fu, Y., He, Y., & Jiao, H. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3306435 · Cited by 9
“Prior works have shown that linear quantization to 8 bits does not cause accuracy loss for 2D-CNNs [46], [47], [48], which is also verified in this work for 3D-CNNs.”
BackgroundInteger Quantized Learned Image Compression
Jeon, G.-W., Yu, S.-E., & Lee, J. (2023) · Proceedings - International Conference on Image Processing · doi:10.1109/icip49359.2023.10222336 · Cited by 15
“While there exist several studies on network quantization for other computer vision and image processing tasks [7], quantization of LIC has not been addressed much.”
BackgroundPruning Meta-Trained Networks for On-Device Adaptation
Gao, D., He, X., Zhou, Z., Tong, Y., & Thiele, L. (2021) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/3459637.3482378 · Cited by 8
“An intuitive remedy for compact models is network pruning, which has the potential to radically remove unimportant parameters in a neural network without deterioration in inference accuracy [5].”
BackgroundWu, L., & Zhang, Z. (2022) · Optics Express · doi:10.1364/oe.462370 · Cited by 9
Hardware-Software Codesign of DNN Accelerators Using Approximate Posit Multipliers
Glint, T., Prasad, K., Dagli, J., Gandhi, K., Gupta, A., Patel, V., Shah, N., & Mekie, J. (2023) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1145/3566097.3567866 · Cited by 8
“2 Background DNN workloads consisting of convolution layers (CL) and fully connected layers (FCL) can be processed using DNN accelerators [1, 2, 16].”
BackgroundMethodInfluentialDGL: Device Generic Latency Model for Neural Architecture Search on Mobile Devices
Wang, Q., & Zhang, S. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3244170 · Cited by 15
“means that NAS need to search for architectures that satisfy the hardware efficiency constraints of devices such as inference latency [5], [6], energy consumption [7], [8], [9], and memory footprint [10], [11].”
BackgroundTraining-aware Low Precision Quantization in Spiking Neural Networks
Shymyrbay, A., Fouda, M. E., & Eltawil, A. M. (2022) · Asilomar Conference on Signals, Systems and Computers (ACSSC) · doi:10.1109/ieeeconf56349.2022.10051957 · Cited by 8
“Having SNN structure as shown in Table I, the full-precision SNN model outperforms other available full-precision models [12], [13], [14], and [20], while the binary model is on par with full-precision models on N-MNIST dataset.”
MethodKlippel, E., Oliveira, R., Maslov, D., Bianchi, A. G. C., Delabrida, S. E., & Garrocho, C. T. B. (2022) · SN Computer Science · doi:10.1007/s42979-022-01169-y · Cited by 8
“The joint execution of these functionalities by the PE allows the implementation of the convolution, polling, and feedforward processes of the deep neural networks [17].”
MethodEliahu, A., Ben-Hur, R., Ronen, R., & Kvatinsky, S. (2020) · IFIP/IEEE International Conference on Very Large Scale Integration · doi:10.1109/vlsi-soc46417.2020.9344103 · Cited by 8
“Numerous accelerators integrating analog memristorbased computations have recently been developed [5].”
BackgroundDeploying Optimized Deep Vision Models for Eyeglasses Detection on Low-Power Platforms
Giedra, H., Sledević, T., & Matuzevičius, D. (2025) · Electronics · doi:10.3390/electronics14142796 · Cited by 7
Autoencoder-based decentralized federated learning for efficient communication
Mamond, A. W., Kundroo, M., & Kim, T. (2025) · Computer Networks · doi:10.1016/j.comnet.2025.111676 · Cited by 7
Lee, K., Hwangbo, S., Yang, D., & Lee, G. (2023) · International Journal of Computational Intelligence Systems · doi:10.1007/s44196-023-00202-z · Cited by 7
“According to [5], research into model compression in DNNs can be classified into four major categories: compact-model methods, tensor decomposition, data quantization, and network sparsification.”
MethodA Low-Power Streaming Speech Enhancement Accelerator for Edge Devices
Wu, C.-H., & Chang, T. (2024) · IEEE Open Journal of Circuits and Systems · doi:10.1109/ojcas.2024.3387849 · Cited by 10
“In the context of domain-aware pruning, most contemporary pruning techniques, as discussed in [11], are tailored for homogeneous architectures.”
BackgroundMethodInfluentialSeatbelt Detection Algorithm Improved with Lightweight Approach and Attention Mechanism
Qiu, L., Rao, J., & Zhao, X. (2024) · Applied Sciences · doi:10.3390/app14083346 · Cited by 7
A Two-Stage ECG Classifier for Decentralized Inferencing Across Edge-Cloud Continuum
Li, X., Cardiff, B., & John, D. (2024) · IEEE Sensors Journal · doi:10.1109/jsen.2024.3406780 · Cited by 8
“However, this introduces fresh concerns regarding: 1) limited computational and memory resources [13] and 2) high power consumption due to the model complexity [14].”
BackgroundA Case for Low Bitwidth Floating Point Arithmetic on FPGA for Transformer Based DNN Inference
Wu, J., Song, M., Zhao, J., & So, H. K. (2024) · IEEE International Symposium on Parallel & Distributed Processing, Workshops and Phd Forum · doi:10.1109/ipdpsw63119.2024.00045 · Cited by 7
“Researchers have found that different parts of DNN models show varying levels of vulnerability to quantization errors [24].”
BackgroundRajanediran, D. K. J., Babu, C. G., & Priyadharsini, K. (2024) · Sustainable Computing Informatics and Systems · doi:10.1016/j.suscom.2024.101042 · Cited by 7
Accurate and Ubiquitous Floor Identification at the Edge using a Single Cell Tower
Mostafa, S., Youssef, M. A., & Harras, K. A. (2024) · IEEE/ACM Symposium on Edge Computing (SEC) · doi:10.1109/sec62691.2024.00024 · Cited by 7
“A possible refinement is to employ model compression techniques [79] to reduce the model size with minimal effects on accuracy.”
AugFL: Augmenting Federated Learning With Pretrained Models
Yue, S., Qin, Z., Deng, Y., Ren, J., Zhang, Y., & Zhang, J. (2025) · IEEE Transactions on Networking · doi:10.1109/ton.2025.3548277 · Cited by 13
“However, compression for large models is still in the early stage and remains highly nontrivial due to the complex configuration and black-box nature thereof [30].”
Yang, T., He, Z., Kou, T., Li, Q., Han, Q., Yu, H., Liu, F., Liang, Y. E., & Jiang, L. (2021) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3467476 · Cited by 9
Accelerating On-Chip Training with Ferroelectric-Based Hybrid Precision Synapse
Luo, Y., Wang, P., & Yu, S. (2022) · ACM Journal on Emerging Technologies in Computing Systems · doi:10.1145/3473461 · Cited by 7
“More details on DNN training can be found in [21].”
BackgroundHigh performance inference of gait recognition models on embedded systems
Ruiz-Barroso, P., Castro, F. M., Delgado-Escaño, R., Cózar, J. R., & Guil, N. (2022) · Sustainable Computing Informatics and Systems · doi:10.1016/j.suscom.2022.100814 · Cited by 9
Pedestrian and Vehicle Detection Based on Pruning YOLOv4 with Cloud-Edge Collaboration
Wang, H., Mo, R., Chen, Y., Lin, W., Xu, M., & Liu, B. (2023) · Computer Modeling in Engineering & Sciences · doi:10.32604/cmes.2023.026910 · Cited by 7
A NoC-Based Spatial DNN Inference Accelerator With Memory-Friendly Dataflow
Zhu, L., Fan, W., Dai, C., Zhou, S., Xue, Y., Lu, Z., Li, L., & Fu, Y. (2023) · IEEE Design and Test · doi:10.1109/mdat.2023.3310199 · Cited by 7
A systematic DNN weight pruning framework based on symmetric accelerated stochastic ADMM
Yuan, M., Bai, J., Jiang, F., & Du, L. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.127327 · Cited by 7
Zills, F., Schäfer, M. R., Tovey, S., Kästner, J., & Holm, C. (2024) · Faraday Discussions · doi:10.1039/d4fd00025k · Cited by 7
Understanding the functional roles of modelling components in spiking neural networks
Yin, H., Zheng, H., Mao, J., Ding, S., Liu, X., Xu, M., Hu, Y., Pei, J., & Deng, L. (2024) · Neuromorphic Computing and Engineering · doi:10.1088/2634-4386/ad6cef · Cited by 9
“This action leads to sparser spike activities, which is advantageous for higher computational efficiency [39, 40].”
Federated Heterogeneous Contrastive Distillation for Molecular Representation Learning
Feng, J., Wang, Z., Wei, Z., Li, Y., Ding, B., & Xu, H. (2024) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/3627673.3679725 · Cited by 7
Wang, H., Guo, B., Zeng, Y., Chen, M., Ding, Y., Zhang, Y., Yao, L., & Yu, Z. (2025) · ACM Transactions on Information Systems · doi:10.1145/3715098 · Cited by 12
Tensor-Train networks for learning predictive modeling of multidimensional data
da Costa, M. N., Attux, R., Cichocki, A. S., & Romano, J. M. T. (2025) · Neurocomputing · doi:10.1016/j.neucom.2025.130037 · Cited by 7
Yang, Y., Chi, X., Deng, L., Yan, T., Gao, F., & Li, G. (2022) · Neurocomputing · doi:10.1016/j.neucom.2022.08.045 · Cited by 10
Self-citationEdge Intelligence for Smart Metro Systems: Architecture and Enabling Technologies
Liu, X., Zhang, M., Zou, C., Yang, J., & Yan, X. (2021) · IEEE Network · doi:10.1109/mnet.211.2100302 · Cited by 8
Deep Learning Approach at the Edge to Detect Iron Ore Type
Klippel, E., Bianchi, A. G. C., Delabrida, S. E., Silva, M. C., Garrocho, C. T. B., Moreira, V. S., & Oliveira, R. (2021) · Sensors · doi:10.3390/s22010169 · Cited by 7
“PEs allow the implementation of the CNN convolution, polling, and feedforward processes [16].”
MethodBhalgaonkar, S., Munot, M. V., & Anuse, A. D. (2022) · Lecture notes in electrical engineering · doi:10.1007/978-981-19-1520-8_55 · Cited by 7
Dai, F., Chen, Y., Huang, Z., Zhang, H., Zhang, H., & Xia, C. (2022) · The Journal of Supercomputing · doi:10.1007/s11227-022-04945-y · Cited by 7
“The value of distributed SRAM can be greatly reduced if we adopt a state-of-the-art pruning technique for the neural network [51].”
MethodLight-Weight EPINET Architecture for Fast Light Field Disparity Estimation
Hassan, A., Sjöström, M., Zhang, T., & Egiazarian, K. (2022) · IEEE International Workshop on Multimedia Signal Processing (MMSP) · doi:10.1109/mmsp55362.2022.9949378 · Cited by 7
“Several strategies have been proposed to reduce the complexity of spatial convolutions [23].”
BackgroundRodríguez, I., Campos, C., & Fdez‐Riverola, F. (2023) · Neurocomputing · doi:10.1016/j.neucom.2023.126835 · Cited by 7
An efficient pruning and fine-tuning method for deep spiking neural network
Meng, L., Qiao, G., Zhang, X. Y., Bai, J., Zuo, Y., Zhou, P., Liu, Y., & Hu, S. (2023) · Applied Intelligence · doi:10.1007/s10489-023-05056-8 · Cited by 11
SASRNet: Slimming-Assisted Deep Residual Network for Image Steganalysis
Huang, S., Zhang, M., Ke, Y., Di, F., & Kong, Y. (2024) · Computing and Informatics · doi:10.31577/cai_2024_2_295 · Cited by 7
A Scatter-and-Gather Spiking Convolutional Neural Network on a Reconfigurable Neuromorphic Hardware
Zou, C., Cui, X., Kuang, Y., Liu, K., Wang, Y., Wang, X., & Huang, R. (2021) · Frontiers in Neuroscience · doi:10.3389/fnins.2021.694170 · Cited by 7
Hardy, E., & Badets, F. (2021) · arXiv · doi:10.48550/arxiv.2103.04792 · Cited by 6
“A comprehensive study of state-of-the-art CNN and RNN quantization examples has been done in [6].”
BackgroundDeep Hybrid Compression Network for Lidar Point Cloud Classification and Segmentation
Zhao, Z. S., Ma, Y., Xu, K., & Wan, J. (2023) · Remote Sensing · doi:10.3390/rs15164015 · Cited by 7
Regret Bounds for Online Learning for Hierarchical Inference
Al-Atat, G., Datta, P., Moharir, S., & Champati, J. P. (2024) · ACM International Symposium on Theory, Algorithmic Foundations, and Protocol Design for Mobile Networks and Mobile Computing (MobiHoc) · doi:10.1145/3641512.3686392 · Cited by 10
Fornt, J., Reggiani, E., Fontova-Musté, P., Rodas, N., Pappalardo, A., Ünsal, O., Cristal, A., Altet, J., Moll, F., & Abella, J. (2024) · IEEE Transactions on Computers · doi:10.1109/tc.2024.3500369 · Cited by 7
“Hardware acceleration of quantized DNNs is a widespread research topic [32].”
Yang, J., & Zou, Y. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.126132 · Cited by 8
Huang, P., Chen, K., Wang, C., & Liu, W. (2025) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2024.3522307 · Cited by 6
“Many applications are inherently error-tolerant, where achieving a totally exact result is unnecessary and leads to excessive consumption of computing resources [1], [2].”
Assessing the Impact of Artificial Intelligence on Healthcare Sector for Enhancing Sustainability
Crowther, D., Singh, R., Singh, A. P., & Rani, S. (2024) · Approaches to global sustainability, markets, and governance · doi:10.1007/978-981-97-9555-0_1 · Cited by 7
Knowledge-guided quantization-aware training for EEG-based emotion recognition
Zhong, S.-H., Shi, J., & Wang, Y. (2025) · Journal of Visual Communication and Image Representation · doi:10.1016/j.jvcir.2025.104415 · Cited by 6
Chen, Z., Xu, D., Shen, C. Q., Ye, Y., & Jiang, B. (2025) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2025.3556081 · Cited by 7
“Second, while DNNs achieve unparal-leled accuracy, this comes at the cost of substantial memory consumption and high computational complexity, which limits their deployment on resource-constrained hardware platforms [8].”
Nayan, M. M. R., Raj, R., Shaik, G. B., Krishna, T., & Naeemi, A. (2025) · Design, Automation, and Test in Europe · doi:10.23919/date64628.2025.10992860 · Cited by 10
“Two Multiplexers (MUX1 and MUX2) are used to select the target buffer based on WS/IS dataflow during preloading.”
Ensemble early exit network on human activity recognition using wearable sensors
Yu, J., Zhang, L., Cheng, D., Bu, C., Liu, L., Wu, H., & Song, A. (2025) · Computer Networks · doi:10.1016/j.comnet.2025.111409 · Cited by 6
Epilepsy Prediction via Time-Frequency Features and Multi-Scale Hybrid Neural Networks
Chang, W., Ji, B., Li, D., Lei, Z., Wei, Y., Liu, X., & Yan, G. (2025) · Journal of Medical Systems · doi:10.1007/s10916-025-02224-w · Cited by 6
“The exponential growth in hardware computational power in recent years is sufficient to support the computational demands of these models [39].”
DECA: A Near-Core LLM Decompression Accelerator Grounded on a 3D Roofline Model
Gerogiannis, G., Eyerman, S., Georganas, E., Heirman, W., & Torrellas, J. (2025) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1145/3725843.3756073 · Cited by 8
Tensor Low-Rank Orthogonal Compression for Convolutional Neural Networks
He, Y., & Luo, X. (2026) · IEEE/CAA Journal of Automatica Sinica · doi:10.1109/jas.2025.125213 · Cited by 6
One-Cycle Structured Pruning via Stability-Driven Subnetwork Search
Ghimire, D., Kil, D., Jeong, S., Park, J., & Kim, S. (2026) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv61042.2026.00530 · Cited by 6
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.