Efficient Neural Network Deployment: A Review of Compression Techniques for Edge Computing
Keywords:
Model Compression, Edge Computing, Neural Network Pruning, Quantization-Aware Training, Knowledge Distillation, Hardware-Aware Optimization, Deep Learning Deployment, Resource-Constrained AIAbstract
This review talks about the growing difference in intelligence between big deep learning architectures and edge hardware that has strict resource limits, like mobile SoCs and IoT sensors. The main goal is to find out how to use different compression methods, like pruning, quantization, and knowledge distillation, without slowing down real-time performance. Our method is based on a systematic look at secondary data from top-tier peer-reviewed research. We compare theoretical accuracy gains with real hardware benchmarks like latency and energy use. The main findings show that while pruning offers high theoretical compression, quantization (especially INT8) is still the best way to save power right away in the industry. Also, it has been shown that hardware-aware optimization and hybrid techniques work much better than isolated methods. We come to the conclusion that there is an urgent need for standardized reporting metrics and policy frameworks to make sure that compressed models are safe and reliable in autonomous systems. This study gives engineers a useful set of tools to help them move AI from the cloud to the edge.
References
Gao, X., Ang, M. C., Althubiti, S. A. (2023). Deep Reinforcement Learning and Markov Decision Problem for Task Offloading in Mobile Edge Computing. Journal of Grid Computing, 21(4), 78. https://doi.org/10.1007/s10723-023-09708-4
Knez, T., Machidon, O., Pejović, V. (2021). Self-Adaptive Approximate Mobile Deep Learning. Electronics, 10(23), 2958. https://doi.org/10.3390/electronics10232958
Li, Q., Zhou, M-T., Ren, T-F., Jiang, C-B., Chen, Y. (2023). Partitioning Multi-layer Edge Network for Neural Network Collaborative Computing. EURASIP Journal on Wireless Communications and Networking, 2023(1), 80. https://doi.org/10.1186/s13638-023-02284-x
Li, W., Hacid, H., Almazrouei, E., Debbah, M. (2023). A Comprehensive Review and a Taxonomy of Edge Machine Learning: Requirements, Paradigms, and Techniques. AI, 4(3), 729. https://doi.org/10.3390/ai4030039
Liu, J., Xiang, J., Jin, Y., Liu, R., Yan, J. (2021). Boost Precision Agriculture with Unmanned Aerial Vehicle Remote Sensing and Edge Intelligence: A Survey. Remote Sensing, 13(21), 4387. https://doi.org/10.3390/rs13214387
Nazir, A., Mir, R. N., Qureshi, S. (2020). Exploring Compression and Parallelization Techniques for Distribution of Deep Neural Networks Over Edge–Fog Continuum- A Review. International Journal of Intelligent Computing and Cybernetics, 13(3), 331-364. https://doi.org/10.1108/IJICC-04-2020-0038
Onteddu, A. R., Rahman, K., Roberts, C., Kundavaram, R. R., & Kothapalli, S. (2022). Blockchain-enhanced machine learning for predictive analytics in precision medicine. Silicon Valley Tech Review, 1(1), 48–60.
Pääkkönen, P., Pakkala, D. (2020). Extending Reference Architecture of Big Data Systems Towards Machine Learning in Edge Computing Environments. Journal of Big Data, 7(1). https://doi.org/10.1186/s40537-020-00303-y
Patsias, V., Amanatidis, P., Karampatzakis, D., Lagkas, T., Michalakopoulou, K. (2023). Task Allocation Methods and Optimization Techniques in Edge Computing: A Systematic Review of the Literature. Future Internet, 15()8, 254. https://doi.org/10.3390/fi15080254
Rahman, K. (2017). Digital platforms in learning and assessment: The coming of age of artificial intelligence in medical checkup. International Journal of Reciprocal Symmetry and Theoretical Physics, 4(1), 1–5.
Downloads
Published
Issue
Section
License
Copyright (c) 2023 American Observer PressBy default, articles published in this journal are available under the journal's standard copyright and access policy. Authors who choose the Open Access option by paying the applicable Open Access Charge will have their articles published under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) License. The applicable license will be clearly indicated on each published article.
