Efficient Acceleration of Deep Learning Inference on Resource-Constrained Edge Devices: A Review
Open accessThe DOI matches its Crossref record; 410 of the 491 references it cites were found in a registry or an index.
Abstract
Successful integration of deep neural networks (DNNs) or deep learning (DL) has resulted in breakthroughs in many areas. However, deploying these highly accurate models for data-driven, learned, automatic, and practical machine learning (ML) solutions to end-user applications remains challenging. DL algorithms are often computationally expensive, power-hungry, and require large memory to process complex and iterative operations of millions of parameters. Hence, training and inference of DL models are typically performed on high-performance computing (HPC) clusters in the cloud. Data transmission to the cloud results in high latency, round-trip delay, security and privacy concerns, and the inability of real-time decisions. Thus, processing on edge devices can significantly reduce cloud transmission cost. Edge devices are end devices closest to the user, such as mobile phones, cyber–physical systems (CPSs), wearables, the Internet of Things (IoT), embedded and autonomous systems, and intelligent sensors. These devices have limited memory, computing resources, and power-handling capability. Therefore, optimization techniques at both the hardware and software levels have been developed to handle the DL deployment efficiently on the edge. Understanding the existing research, challenges, and opportunities is fundamental to leveraging the next generation of edge devices with artificial intelligence (AI) capability. Mainly, four research directions have been pursued for efficient DL inference on edge devices: 1) novel DL architecture and algorithm design; 2) optimization of existing DL methods; 3) development of algorithm–hardware codesign; and 4) efficient accelerator design for DL deployment. This article focuses on surveying each of the four research directions, providing a comprehensive review of the state-of-the-art tools and techniques for efficient edge inference.
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
“These data cannot be managed by relying only on the limited functional services provided by cloud service centers (Ge et al. 2023;Shuvo et al. 2022).”
From Survey: federated learning data security and privacy-preserving in edge-Internet of Things (2024) · cites this paper as Shuvo et al. 2022
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
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148
ImageNet 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
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173 · cited 1×
“NLP models, such as Transformer [45] and BERT [224], have around 80% [225] and 93% [226] weight sparsity.”
MethodGradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805
You Only Look Once: Unified, Real-Time Object Detection
Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.91 · Cited by 53,803 · cited 1×
“Thus, splitting the computation of convolution and dense layer between GPU and CPU can achieve a 60× speedup for object detection using YOLO in the Jetson TK1 board [283].”
MethodDensely 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
Squeeze-and-Excitation Networks
Hu, J., Shen, L., & Sun, G. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00745 · Cited by 35,667
(2014) · Cited by 33,419
Research works printed but not linked: 30
- Wendelken, 2022, Worldwide and U.S. IoT Cellular Connections Forecast, 2021–2025
- Gupta, ProtoNN: Compressed and accurate KNN for resource-scarce devices, 1331
- Dennis, 2020, EdgeML: Machine learning for resource-constrained edge devices
- Kumar, Resource-efficient machine learning in 2 KB RAM for the Internet of Things, 70, 1935
- 2022, AutoML
- 2022, AutoKeras
- Ryu, BitBlade: Area and energy-efficient precision-scalable neural network accelerator with bitwise summation, 1
- Hojabr, SkippyNN: An embedded stochastic-computing accelerator for convolutional neural networks, 1
- 2022, Jetson TX2 Module
- 2022, Intel® Edison Development Platform
- 2022, Open Neural Network Exchange
- 2022, Deep Learning HDL Toolbox
- 2022, X-CUBE-AI, AI Expansion Pack for STM32CubeMX
- 2022, DNNDK User Guide
- 2022, NVIDIA TensorRT
- 2022, CEVA Deep Neural Network (CDNN)
- 2021, Qualcomm Neural Processing SDK for AI
- 2022, Stratus High-Level Synthesis
- 2022, Deploy Machine Learning Models on Mobile and IoT Devices
- Rasley, DeepSpeed, 3505
- Baines, 2021, FairScale: A general purpose modular PyTorch library for high performance and large scale training
- Dai, BigDL, 50
- 2022, Tensilica DNA Processor Family for on-Device AI
- 2022, DesignWare ARC EV Processors for Embedded Vision
- 2022, AI for the Edge
- 2022, Products, Helping You Bring Local AI to Applications From Prototype to Production
- Shacham, 2022, Pixel Visual Core: Image Processing and Machine Learning on Pixel 2
- Yang, Sparse ReRAM engine: Joint exploration of activation and weight sparsity in compressed neural networks, 236
- Mao, A privacy-preserving deep learning approach for face recognition with edge computing, 1
- Wang, Not just privacy, 2407
Cited by
Items 301–400 of 420
Energy Efficient and QoS-Aware Model Selection for DNN Inference in Edge Intelligence
Siar, H., & Elmroth, E. (2025) · doi:10.1109/cloudcom67567.2025.11331487
“Some works incorporate model selection with partitioning and offloading techniques [4].”
InfluentialMa, C., Feng, Y., Luan, S., Zhao, Z., Bai, Y., Sun, L., Liu, H., & Bo, J. (2026) · IEEE Transactions on Consumer Electronics · doi:10.1109/tce.2026.3655427
“To overcome such di ffi culties, lightweight convolutional neural networks (CNNs) have been trained to be directly executed on the edge devices [24], which are faster in their inference but are not scalable, and cannot evaluate cross-institutional results.”
NeuralTimer: Configuration-Based Neural Network Approach to Hardware Timer-Based Applications
Akmandor, M. O., & Sarioglu, B. (2026) · IEEE Access · doi:10.1109/access.2026.3656830
“…making local embedded intelligence a fundamental requirement rather than an optional enhancement and adaptability The associate editor coordinating the review of this manuscript and approving it for publication was Sotirios Goudos . become as critical as connectivity itself [7], [8], [9], [10].”
Jha, S. K., & Singh, A. P. (2026) · Lecture notes in networks and systems · doi:10.1007/978-981-95-3701-3_46
AI-enhanced ECG analysis for arrhythmia classification with edge device compatibility
Nugroho, H., Peng, K. J., & Shanmugam, S. A. (2026) · Advances in Computational Intelligence · doi:10.1007/s43674-026-00087-3
“The application automatically optimizes runtime parameters including batch size and number of processing streams to achieve optimal performance for each target device (Shuvo et al. 2023).”
SPICE: Structured Pruning for Inference on Constrained Edge Devices
Das, S., Cheng, J., Rakshit, A., Soni, B., Boubin, J. G., & Ramnath, R. (2026) · IEEE Consumer Communications and Networking Conference · doi:10.1109/ccnc65079.2026.11366473 · Cited by 1
“These statistics were obtained using torchinfo with a dummy input of size [1, 3, 576, 576] , matching our original image transformation.”
Self-citationA comparative study of dimensionality reduction techniques for intrusion detection in IoT networks
García-Merino, J. C., Tobarra, L., Robles-Gómez, A., Pastor-Vargas, R., & Sarraipa, J. (2026) · PeerJ Computer Science · doi:10.7717/peerj-cs.3553 · Cited by 1
Lakshmi, A. V., Kumari, T. P., Venkatesh, S., & Goverdhan, P. (2025) · International Conference on Innovations and Emerging Technologies in AI & Communication Systems (IETACS) · doi:10.1109/ietacs68750.2025.11385755
“As the world experiences an outbreak of chronic diseases, the ageing of populations, and as non-invasive, real-time health diagnostics become highly demanded, intelligent biomedical systems capable of robust operation in or on the human body over long durations are urgently required [2].”
Mounika, K., Krishnaveni, S., Renuka, N., & Kalpana, G. (2025) · International Conference on Innovations and Emerging Technologies in AI & Communication Systems (IETACS) · doi:10.1109/ietacs68750.2025.11385734
“Nevertheless, advanced AI models at the edge are yet to be deployed because of the aggressive power, area, and processing latency limitations [2].”
HOLI-SRNet: lightweight reconstruction of super-resolution images on resource-constrained platforms
Ju, Y., Jiang, W. Y., Wang, F., Ji, N., & Liu, Y. (2026) · Journal on Image and Video Processing · doi:10.1186/s13640-026-00689-4
Zhou, Y., Gao, Z., Shao, C., Liang, J., Li, H., & Huang, J. (2025) · IEEE International Conference on Robotics and Biomimetics (ROBIO) · doi:10.1109/robio66223.2025.11376129
SteadyFlow-Edge: Adaptive Frame-Dropping for Stable Real-Time Detection on Edge
He, S.-R., Chen, C., Zheng, Q., et al. (2025) · International Conference on Advances in Computing and Artificial Intelligence · doi:10.1109/acai68217.2025.11406678
“With the continuous development of edge computing and embedded systems, the demand for edge inference systems in practical applications has been steadily increasing [1] [2].”
Gundam: A Generalized Unified Design and Analysis Model for Matrix Multiplication on Edge
Cheng, Q., Li, H., Dong, W.-K., Huang, M., Lin, L., & Hashimoto, M. (2026) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/asp-dac66049.2026.11420712
“However, matrix multiplication also constitutes one of the most resource-intensive operations in AI inference and training, accounting for the majority of computational cost and energy consumption [3], [4].”
IceEdge: Thermal-Aware Machine Learning Inference Serving for Emerging Edge Applications
Pandith, O. S., Pandit, M. K., Gujarati, A., Mehta, A., & Sen, R. (2026) · ACM Transactions on Sensor Networks · doi:10.1145/3801093 · Cited by 1
“The recent developments in ML research has led to deployment of such powerful models on resource constrained edge platforms [30, 31, 37].”
LaPOD: Latency Prediction for Real-Time LiDAR Object Detection
Xie, W., Ren, T., Wu, J., Xue, C. J., & Guan, N. (2026) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/asp-dac66049.2026.11420467
“Latency prediction plays a critical role in real-time perception systems [4], [5], as it enables the estimation of inference latency before the actual execution based on sensor-generated point cloud inputs.”
Object Detection for Smart Retail Systems Using OpenCV and DeepLearning
Yarrangari, S. C., Sharif, M. S., Kure, H. I., et al. (2025) · International Conference on Innovation and Intelligence for Informatics, Computing, and Technologies (3ICT) · doi:10.1109/3ict68299.2025.11442075
Li, L., Si, G., Wang, Z., et al. (2025) · International Conference on Clean Energy Storage and Power Engineering (CESPE) · doi:10.1109/cespe68033.2025.00111
“However, reward-driven optimization and rapid adaptation require systematic integration into small-sample industrial equipment quality assessment [7,8]. encoder to capture domain-level structures and extract local contextual cues from multimodal inputs.”
Efficient Aggregation Based on Straggling Communication in Federated Learning
Liu, Y., & Hou, P. (2026) · Lecture notes in computer science · doi:10.1007/978-981-95-8411-6_32
Li, Y., Shen, C., Zhan, W., Xu, B., Yang, Y., Zhang, X., Wang, H., Zhao, C., & Jian, H. (2026) · Journal of Semiconductors · doi:10.1088/1674-4926/25070029
TinyML: Applications, Algorithms, Co-design and Implementations
Ghazal, O. G., Rahman, T., & Shafik, R. (2026) · Springer eBooks · doi:10.1007/978-3-032-06286-4_6
Zhou, H., Lu, L., Yang, H., & Zhang, Y. (2026) · Computers · doi:10.3390/computers15040223
Accelerated Application of AI Model Lightweight Technology in Edge Computing
Zhou, Z. D., Wang, W., Xu, M., & Sun, Y. (2026) · Smart innovation, systems and technologies · doi:10.1007/978-3-032-09601-2_6
Lamaamar, O., Farhan, Y., & Madi, A. A. (2025) · International Conference on Advanced Technologies and Interdisciplinary Innovation (ICAT2I) · doi:10.1109/icat2i69744.2025.11472779
“As is confirmed in these studies [9], [10], the main key of AI edge device is to achieve an optimal tradeoffs among accuracy, performance system, low cost and silicon area.”
Learning to Cascade: A Pomdp Approach to Sequential Model Selection
Darjani, A. S., & Zois, D. (2026) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp55912.2026.11464865
Guo, C., Li, Y. H., Ma, J., Fang, Z., Niu, Z., & Xu, H. (2026) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp55912.2026.11464428
TinyML for Edge Devices: Current State, Technological Advancements, and Future Opportunities
Lamba, A. K., Pal, D., Koli, S., Yamsani, N., & Saini, A. (2026) · International Conference on Advancing Sustainable Solutions through Technologies (ICASST) · doi:10.1109/icasst68917.2026.11484504
“In addition, edge devices need to be capable of bootstrapping AI workloads on minimal instruction sets, which means that they do not have floating-point arithmetic units[9].”
Energy- and Quantization-aware DNN Partitioning in the Edge-Cloud Continuum (Work In Progress Paper)
Nicosanti, S., Russo, G. R., & Cardellini, V. (2026) · ACM/SPEC International Conference on Performance Engineering (ICPE) · doi:10.1145/3777911.3801106
Towards Object Counting by Efficient Scale Recognition Network
Zhai, W., Zhang, T., Zhang, T., Fan, Z., Gao, J., & Xing, X. (2025) · China Automation Congress (CAC) · doi:10.1109/cac67268.2025.11487391
“Such models, though highly accurate, face practical limitations when applied to edge devices with limited resources [8, 9].”
Optimal DNN Partitioning and Deployment for Dependable Inference Workflow on End Devices
Liu, J., Wang, X. W., Fei, S., et al. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3688449
“Compared to EEC inference architectures, end devices equipped with CPU-FPGA heterogeneous computing platforms offer significant cost-effectiveness advantages for DNN inference tasks [6].”
Fu, Y., Qin, P. D., & Jiang, C. (2026) · IEEE Communications Surveys & Tutorials · doi:10.1109/comst.2026.3690731 · Cited by 4
“Edge-compatible models can be designed through four-level adaptation, i.e., model-level [21], depth-level [84], parameter-level [85], and bit-level [86], with the aim of shrinking model size while saving computation and memory footprint.”
InfluentialA phase-adaptive weighted approach for self-data-driven RUL prediction of machinery
Fang, R., Gao, Y., Lei, Y., Yang, B., Li, X. L., & Li, N. (2026) · Measurement Science and Technology · doi:10.1088/1361-6501/ae693c
“…requiring extensive and fully labeled historical failure datasets for training, which are often unavailable for specific industrial assets; second, their complex network structures involve numerous parameters, resulting in considerable computational latency and memory overhead [7, 8].”
A Deployment-Oriented Extreme Learning Machine for Electricity Price Forecasting
Elashaal, M., Shohag, M. S. A., & Wu, Q. M. J. (2025) · International Conference on Computer and Information Technology (ICCIT) · doi:10.1109/iccit68739.2025.11491138
“Similar to other domains [14], [15], deep learning models used in EPF are also memory-intensive and computationally expensive, which makes their efficient deployment on embedded edge devices challenging [16], [17].”
Emmanuel, D. E., & Fernando, A. (2026) · Research Square · doi:10.21203/rs.3.rs-9542617/v1
Smart IoT-Integrated Multi-Device Health Monitoring Framework with AI-Driven Predictive Alert System
Vignesh, N., Reddy, G. B. S. N., Jagadeesh, C., et al. (2025) · International Conference on Sustainable Computing and Smart Systems (ICSCSS) · doi:10.1109/icscss64956.2025.11501036
“Furthermore, many models lack the computational efficiency required for real-time deployment on resource-constrained edge devices, such as wearable platforms or embedded healthcare gateways [9], [10].”
Transforming VLSI Design with AI: Pioneering the Future of Chip Technology
Jyothy, A., & John, A. T. (2025) · International Journal of Web Technology · doi:10.20894/ijwt.104.014.001.005
Data-driven subgrid-scale turbulence parameterization: methods for improving model efficiency
Zhu, X., Qu, Y., Wu, N., Zhang, H., & Shi, X. (2026) · Machine Learning Earth · doi:10.1088/3049-4753/ae6dbe
“It should be mentioned that FLOPs only objectively describe the computational complexity of a DL model, but the actual cost of its inference time is influenced by many other factors, such as the memory access cost, memory transfer rate, hardware, parallelization algorithm and so on [22, 31].”
Efficient Human Pose Estimation for Edge Devices via Structure-Aware and Dynamic Modeling
Li, T., & Liu, S. (2026) · International Conference on Computer, Artificial Intelligence and Control Engineering (CAICE) · doi:10.1145/3804601.3804713
“However, edge devices are severely limited in computational power, storage, and energy [12], making traditional high-performance models unsuitable for direct deployment.”
Han, Y., Bi, Q., Xiang, S., ZHANG, Y., Guo, X., Li, J. H., Shi, Y., Pan, W., & Hao, Y. (2026) · Optics Express · doi:10.1364/oe.592163
Reconfigurable VLSI Architectures for Resource-Constrained Edge AI
Girish Gowtham, A., & Subramanian, S. (2026) · International Conference on Artificial Intelligence and Machine Learning Applications Theme: Healthcare and Internet of Things (AIMLA) · doi:10.1109/aimla67915.2026.11522394
“This trend toward localized, high-speed execution is a core theme in the comprehensive review by Shuvo et al. [4], which highlights the severe energy and memory challenges faced when deploying deep learning inference on resource-constrained IoT devices.”
Crop Disease Classification Using Lightweight Deep Learning Models
Chowdhury, A. W., Udoh, E., Gashi, E., et al. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-26211-0_10
Design and Optimization of Low-Power AI Accelerators for Edge Devices
Harshitha, L., B, L. R. U., Vaishnavi, Y., R, J. H., Bindu, S., & Rekha, P. (2026) · International Conference on Knowledge Engineering and Communication Systems (ICKECS) · doi:10.1109/ickecs70176.2026.11527502
“Literature suggests 20– 40% sparsity maintains model performance while enabling measurable efficiency gains [11, 16].”
Wan, X., Li, D., Fu, D., & Wen, D. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3697898
“Post-hoc compression often leads to accuracy degradation for sparse low-density sEMG signals [36], while native lightweight architectures such as TCNs achieve 12.8 ms latency on IoT processors but cannot capture global dependencies [25].”
InfluentialKhaldy, M. A. A., Otaiwi, Z., & Al–Qerem, A. (2026) · International Conference on Computational Intelligence Approaches and Applications (ICCIAA) · doi:10.1109/icciaa68481.2026.11543764
“Surveys on acceleration stress that kernel coverage and compiler transformations determine real speed [4].”
InfluentialAdaptive Privacy Optimization for Energy-Constrained Tiny Federated Learning Devices
Aroussi, S. A., Silkan, H., Ouadrhiri, A. E., & Hadni, M. (2026) · International Conference on Innovative Research in Applied Science, Engineering and Technology (IRASET) · doi:10.1109/iraset68627.2026.11538879
“Offload-ing sensed data to centralized servers is frequently impractical due to high communication costs and data sensitivity; consequently, privacy-preserving edge analytics have become critically important [2], [3].”
Energy-Efficient Deep Learning Models for IoT Edge Devices
Nagarajan, S. K. S., Sristi, J., Tamrakar, G., et al. (2026) · International Conference on Recent Trends in Computer Science and Technology (ICRTCST) · doi:10.1109/icrtcst68392.2026.11545215
RetractedAleassa, O., Bekhit, M., Hussain, W., & Fattah, I. M. R. (2026) · Renewable and Sustainable Energy Reviews · doi:10.1016/j.rser.2026.117171 · Cited by 1
Cheng, Q., Lin, L., Huang, M., Li, Q., Yang, Z., Dai, L., Yu, H., Zhang, R., Chen, Y.-J., Shi, Y., et al. (2026) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2026.3701365
“Neural networks (NNs) can easily reach tens of millions of weight data, leading to a huge memory footprint in real-time applications [1], [2], [3].”
Olarewaju, S., Sadiq, A. S., Kaiwartya, O., & Konios, A. (2026) · Journal of Cybersecurity and Privacy · doi:10.3390/jcp6030103
“Shuvo et al. (2023) [212] observed that deep learning models are “computa-tionally expensive, power-hungry, and require large memory,” with inference ordinarily offloaded to high-performance cloud infrastructure, a model that introduces unacceptable round-trip latency in access-control scenarios…”
A Parameter-Efficient Cascaded Framework for Mosquito Species Identification using MFCC Features
Kiriwanna, N., Nukoolkit, C., & Triyason, T. (2026) · doi:10.1145/3816713.3818800
“The processing requirements of these advanced models make them unsuitable for resource-constrained edge devices in the field [20].”
Han, F., Yan, D., Wang, S., Huang, J., Feng, Y., & Yang, P. (2026) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3810214 · Cited by 1
“In addition, emerging hardware accelerators, software-hardware codesign structures, and edge-assisted deployment paradigms provide promising opportunities to further reduce latency [30, 41].”
RACP: An Efficient RISC-V Domain-Specific Processor for Arbitrary-Size Kernel CNNs
Jiang, B., Ding, J., Lu, T., et al. (2026) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas66217.2026.11562891
Sensor Fusion-Based Smart Glove for Deterministic Sign Language Recognition: An IoT-Enabled System
Ortiz, L. A. P., Sánchez, A. C., Cajamarca, B. G. L., et al. (2026) · Technologies · doi:10.3390/technologies14060371
“Even when edge inference is technically feasible, efficient embedded deployment remains a significant design challenge [15].”
CT-LNN: FPGA-Based Accelerator for Continuous-Time Liquid Neural Networks
Wang, L., Gong, Y., Xi, L., Wang, Y., Wang, Y., & Liu, W. (2026) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2026.3704401
“Hybrid accelerators that combine CNN and RNN / LSTM components must manage heterogeneous computational patterns, which can lead to increased control complexity and potential underutilization of hardware resources [35], [36].”
Cheng, G., Guo, Z., Yan, Z., et al. (2026) · Physical review. B./Physical review. B · doi:10.1103/53wd-bg7k
Ng, D., Lam, K. H., Bu, S. Q., Lo, W. C., Chan, C. H., Ng, R., Chan, S., Mak, M., Wong, H., Chim, S., et al. (2026) · Journal of Low Power Electronics and Applications · doi:10.3390/jlpea16030022
“Given that edge devices are inherently constrained in terms of power, memory, and thermal capacity, these inefficiencies pose significant obstacles to the deployment of large-scale AI models at the network edge [2,4,7].”
InfluentialMultimodal Emotion Recognition Based on STM32 Microcontroller
Jia, Z., Li, Y., Cao, G., & Zhang, J. (2025) · doi:10.1109/ccat68690.2025.00027
“…statically assigning buffers within the 192KB SRAM of STM32F407ZET6; (2) Cortex-M4 core-specific optimizations leveraging CMSIS-NN library [18] to accelerate fixed-point matrix operations for convolutional and fully-connected layers; (3) Generation of complete project templates…”
Ren, T., Yao, Y., Hu, Z., & Niu, J. (2026) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom59046.2026.11571717 · Cited by 1
“Distributed edge inference aims to mitigate cloud latency and bandwidth bottlenecks [34].”
Wang, A., Yuan, X., Kang, H., et al. (2026) · doi:10.1109/infocom59046.2026.11571764
“Such overhead makes these models difficult to deploy on resource-constrained edge devices [4].”
Kim, H., Lee, J., Choi, M., Lee, J., & Cho, J. (2026) · Lecture notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering · doi:10.1007/978-3-032-22500-9_24
Wu, Q., Xu, Z., Zheng, M., et al. (2026) · IEEE Sensors Journal · doi:10.1109/jsen.2026.3707745
“The superiority of the proposed method was demonstrated by comparing it with seven state-of-the-art models, including CNN [21], MK-ResCNN [9], LiConvFormer [22], WKN [13], SWKAN [18], SincMCN [8], and MCN [16].”
AI-Assisted IoT Firefighting Robot for Intelligent Fire Detection and Real-Time Response
Alice Merina, F., Grace Maria, J., Dharani, T., Isha, T., & Sridharan, S. (2026) · International Conference on Electrical and Electronics for Sustainable Innovations (ICEESI) · doi:10.1109/iceesi68020.2026.11583077
Wu, Q., Xie, Q., Zheng, Z., et al. (2026) · Industry science and engineering. · doi:10.62381/i265402
Take a Deep Look at Network Traffic Analysis Based Internet of Things Device Fingerprinting
Zou, Q., Li, Q., Zheng, L., Li, R., Zhang, Y., & Jiang, Y. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3712070
“Several studies have highlighted this accuracy–resource trade-off in IoT fingerprinting and related edge settings [36], [89].”
PCF: Proactive Communication Framework for Edge Collaborative Perception
Yang, Y., Wang, Y., & Nguyen, G. T. K. (2026) · International Conference on Communications · doi:10.1109/icc59461.2026.11587234
“Transmitting all camera streams can oversubscribe the link, increase queuing delays, and degrade inference accuracy as updates become stale or dropped [8], [9].”
A lightweight attention-enhanced framework for crop pest and disease detection in desert greenhouses
Liang, Y., Ouyang, K., Luo, Z., et al. (2026) · Frontiers in Plant Science · doi:10.3389/fpls.2026.1830048
MobileSAM2: Lightweight Segment Anything for Spatial Intelligence
Jiang, K., Huang, J., Zhang, J., Xie, W., Li, Y., Wang, Y., Xiao, A., & Tao, D. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-37032-7_27
Tan, W., Zhang, A., & Xie, G. (2026) · IEEE International Conference on Electronic Technology, Communication and Information (ICETCI) · doi:10.1109/icetci68937.2026.11610829
“On the preprocessing side, traditional techniques like CLAHE enhance contrast but can amplify high-frequency noise when applied in isolation [13] , while learning-based enhancement methods introduce prohibitive overhead violating real-time constraints [14] .”
InfluentialGuo, H., Zhang, X., Wang, X., Fei, J., Li, J., Du, Y., & Cao, J. (2026) · Cluster Computing · doi:10.1007/s10586-026-06343-y
“In this method, forward and backward propagation operations are independently performed on worker nodes [21].”
Liu, Q., Luo, S., & Zhan, Y. (2026) · International Conference on Computer Intelligence and Software Engineering (CICSE) · doi:10.1109/cicse70286.2026.11620993
“The cloud stored the sample library, policy version and model parameters for offline training, staged update and log audit [2].”
Thorat, V., & Kshirsagar, D. D. (2026) · International Conference on Intelligent Technologies (CONIT) · doi:10.1109/conit69683.2026.11621509
““Lin and Wang [11] looked at edge-device optimization for restoration pipelines, while Yamamoto et al. [12] investigated GPU accelerated real-time enhancement frameworks.”
Intelligent chemical sensors: learning-enabled platforms for adaptive chemical detection
Batool, R., Sonil, N. I., Nazar, M. F., & Ullah, Z. (2026) · RSC Advances · doi:10.1039/d6ra05215k · Cited by 2
“13,27,45 Fully adaptive intelligent chemical sensors would integrate responsive materials, recon gurable device operation, online learning, embedded decision-making, and closed-loop feedback to sustain reliable performance in real-world environments.”
InfluentialCoral: Covariance-Guided Resource Adaptive Learning for Efficient Edge Inference
Nabhaan, A. N. L., Sukma, Z., Rachmanto, R. D., et al. (2026) · IEEE International Conference on Fog and Edge Computing (ICFEC) · doi:10.1109/icfec69006.2026.00012
“Real-world deployments of edge applications, e.g. , from autonomous drones to smart surveillance, demand consistent inference throughput under strict power constraints [1].”
Optimizing Deep Learning Architectures for the Device-Edge-Cloud Continuum in IoT
Vijayakumar, K., & Munusamy, E. (2026) · Advances in computational intelligence and robotics book series · doi:10.4018/979-8-2600-0615-3.ch007
Luo, H., Zhong, Q., Guo, Q., & Li, X. (2026) · Scientific Reports · doi:10.1038/s41598-026-64451-6
“14, 15, 18, 19 and the corresponding tables, the pruning not only does not impair the model performance, but also brings a slight improvement in accuracy, which is consistent with the The Lottery Ticket Hypothesis 43 : there are a large number of redundant parameters in the original large model,…”
Khan, A., Junaid, A., Iqbal, A., Husnain, G., Shin, W., & Lim, S. (2026) · IEEE Access · doi:10.1109/access.2026.3720106 · Cited by 1
“Compilation settings should specify whether compilation is enabled, whether graph capture is used and whether kernel fusion applies under the chosen runtime [133].”
Ouyang, Y., Chen, L., Wang, Q., & Li, J. (2026) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2026.3717165
“This limitation is further exacerbated in production inference deployments [10], [11], [12].”
Cui, L., Guo, Y., 彭纬璇, & Huang, Y. (2026) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/2575-8411.2026.00134
“Some studies have proposed solutions using FPGA (Field Programmable Gate Array) [34] and other specialized hardware.”
Farahmand, E., Soumma, S. B., Chatrudi, N. T., & Ghasemzadeh, H. G. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3683774 · Cited by 11
“Our GlucoNet with KD achieves high accuracy with only about 10,900 total parameters, which makes it suitable for potential edge deployment [54].”
Ji, C., Li, X., Lv, Z., Sun, H., Liu, P., Yang, H. F., & Pu, Z. (2026) · Engineering · doi:10.1016/j.eng.2026.07.031
An NPU-Accelerated Edge Computing System for Real-Time Underwater Fish Detection
Liu, F., Geng, P., Cui, K., et al. (2026) · doi:10.1109/oceans66983.2026.11616926
“To ensure long-term functionality, these devices must meet stringent low-power requirements [9], [10].”
A Comprehensive Survey on Embedded Deep Learning Algorithms for Smart IoT Monitoring Systems
Alalwany, W. S. H., & Fahim, A. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-25568-6_9
Chaotic Dynamic Walrus Optimization for Efficient Brain Tumor Classification in the Neurojico Era
Askr, H., Moussa, F., Tealeb, A., et al. (2026) · International Journal of Computational Intelligence Systems · doi:10.1007/s44196-026-01376-y
Generative AI-driven sensor fusion for missing-modality and open-set smartphone activity recognition
Shdefat, A. Y., Amine, J. A. E., Α, E. E., Elsayed, F., & Moustafa, N. (2026) · Discover Computing · doi:10.1007/s10791-026-10408-2
“The similarity analysis supports this interpretation, since the synthesized signals show strong correlation and spectral agreement with the real signals while maintaining low MSE and DTW distance [35].”
Water Quality Monitoring Using IoT: Innovations and Implementation
Salman, N. A., Al-Saad, H. T., Al-Mishrey, M. K., & Rushdi, A. I. (2026) · Springer eBooks · doi:10.1007/978-981-92-0294-2_11
QROOT: Quantum-Assisted Root Cause Analysis for Enhanced Observability in Microservice Architectures
shin, S., & Kim, Y. (2026) · IEEE Access · doi:10.1109/access.2026.3723579
Aishwarya, J., Karan, R., Barikai, P. H., et al. (2026) · International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE) · doi:10.1109/icdcece68961.2026.11645786
“For reducing translation time, especially when it comes to performing complicated tasks for constrained edge devices, hardware-based mechanisms known as Translation Lookaside Buffers (TLBs) serve as crucial caches for keeping track of recent virtual-to-physical address translations [11].”
EVA-NAS: Energy Variance Aware Neural Architecture Search for Resource Constrained Embedded Systems
Yousset, H. A., Raad, M., Valle, M., et al. (2026) · doi:10.1109/smacd70206.2026.11647839
“The rapid growth of embedded intelligent systems has driven increasing demand for efficient deep learning models capable of operating under strict resource constraints [1].”
Wijaya, I. G. N. A. M., & Sidqi, M. A. (2026) · International Conference on Smart Mechatronics (ICSMech) · doi:10.1109/icsmech69390.2026.11647114
“Localized edge-device deployment strictly dictates sub-second execution limits, typically less than 100 ms, alongside severely constrained physical memory tolerances [24], [25].”
Min, J.-H., Bhat, A., Assoa, A. S., 조서영, Kim, J., & Ryu, S. (2026) · International Conference on Synthesis, Modeling, Analysis and Simulation Methods and Applications to Circuit Design · doi:10.1109/smacd70206.2026.11647790
Empirical Investigation of Adaptive Machine Learning Pipelines for Resource-Constrained Applications
Parameshwarappa, N., Chakravarty, S., Prasad, U., et al. (2026) · doi:10.1109/icndsa68777.2026.11652225
“This will have the effect of reducing the number of non-zero parameters since it basically decreases the memory bandwidth required [13].”
Design Space Exploration of Convolutional Neural Network Accelerators for Predictive Maintenance
Rokohl, F., Mondal, S., Farsa, E. Z., et al. (2026) · IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW) · doi:10.1109/ipdpsw71298.2026.00055
“However, edge computing addresses latency, privacy, and connectivity concerns by processing data locally [3].”
Chen, G., Sun, S., Wang, X., et al. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7322367
Budgeted Neuro-Symbolic Decision for Industrial Online Repair Under Expensive Verification
Jiang, H., You, J., Chen, Z., Ming, X., & Sun, P. Z. H. (2026) · IEEE Transactions on Automation Science and Engineering · doi:10.1109/tase.2026.3728052
Lamaakal, I., Yahyati, C., Maleh, Y., Makkaoui, K. E., & Ouahbi, I. (2026) · Internet of Things · doi:10.1016/j.iot.2026.102082 · Cited by 1
A Hybrid Method for Semantic Cache Analysis of Computational Kernels in C Programs
Egunov, V. A., Kravets, A. G., Kravchenya, P. D., Матохина, А., & Shabalovsky, V. A. (2026) · Applied System Innovation · doi:10.3390/asi9090177
Orchestrating Composite Applications at the Edge
Calagna, A., Ravera, S., & Chiasserini, C. F. (2026) · IEEE Network Operations and Management Symposium (NOMS) · doi:10.1109/noms69089.2026.11668171
Self-Evolving Neural Systems for Heterogeneous Mobile Devices
Li, Z., Zhang, P., Yu, E., Dai, H., He, Q., & Liu, S. (2026) · IEEE International Conference on Web Services (ICWS) · doi:10.1109/icws72778.2026.00091
Khan, M. A., Alkhattabi, M., Shah, S. F. A., Khan, M. A., Mazhar, T., Saudagar, A. K. J., Rehman, A. U., Hamam, H., & Cui, J. (2026) · Chinese Journal of Electronics · doi:10.23919/cje.2025.00.347
ORFC: Orthogonal Reparameterization for Low-Bitrate ViT Feature Coding
Zhang, L., Yang, W., & Duan, L. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-37356-4_11
Metadata
- Authors
- Shuvo, M. M. H., Islam, S. K., Cheng, J., & Morshed, B. I.
- Affiliations
- University of Missouri (US)
- Texas Tech University (US)
- Venue
- Proceedings of the IEEE
- Published
- 2023-01
- Volume
- 111
- Issue
- 1
- Pages
- 42-91
- ISSN
- 0018-9219 (ISSN-L)
- 1558-2256
- OpenAlex
- W4312910656
- Cited by
- 436
- References
- 491
- Citations by year
- 2026: 159 · 2025: 173 · 2024: 82 · 2023: 22
- FWCI
- 22.67
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Locations
- Proceedings of the IEEE (publishedVersion, cc-by)
- ThinkTech (Texas Tech University) (submittedVersion, cc-by)
- ThinkTech (Texas Tech University) (submittedVersion)
- Funders
- College of Engineering, University of Missouri, Columbia, MO, USA
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
- Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
- Keywords
- deep learning inference, edge computing, resource-constrained devices, inference acceleration, algorithm-hardware co-design, neural network accelerators, model compression, Internet of Things, embedded systems, energy-efficient computing
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.