Neural architecture search for optimizing edge computing in IoT devices

  • Authors

    • Dr. Lydia Languish School of Business, University of Sydney, Australia. Author

    DOI:

    https://doi.org/10.67228/30713315/IJAIDT-2021PI3S4N

    Published 03-04-2021

  • Neural Architecture Search (NAS), Edge Computing, Internet of Things (IoT), Model Optimization, Resource-Constrained Devices, Hardware-aware NAS, Energy Efficiency, Deep Learning

    Issue

    Vol. 4 No. 1 (2021)

    Section

    Articles

    How to Cite

    [1]
    L. Languish, “Neural architecture search for optimizing edge computing in IoT devices”, IJAIDT, vol. 4, no. 1, pp. 01–12, Mar. 2021, doi: 10.67228/30713315/IJAIDT-2021PI3S4N.

  • Abstract

    The proliferation of Internet of Things (IoT) devices has intensified the demand for efficient and accurate deep learning models capable of operating under stringent resource constraints at the edge. Neural Architecture Search (NAS) offers a promising avenue to automate the design of optimized neural networks tailored for edge computing environments. This paper investigates the application of NAS for optimizing neural network architectures deployed on IoT edge devices, balancing accuracy, latency, and energy efficiency. We propose a multi-objective NAS framework that incorporates hardware-aware constraints specific to typical IoT edge platforms. Experimental results on benchmark datasets demonstrate that NAS-generated models outperform conventional architectures in terms of inference speed and power consumption, while maintaining competitive accuracy. Our findings highlight the potential of NAS as a vital tool for enhancing edge intelligence in IoT systems.

  • References

    [1] Zoph, B., & Le, Q. V. (2017). Neural architecture search with reinforcement learning. arXiv preprint arXiv:1611.01578.

    [2] Tan, M., & Le, Q. (2019). EfficientNet: Rethinking model scaling for convolutional neural networks. ICML.

    [3] Han, S., Pool, J., Tran, J., & Dally, W. J. (2015). Learning both weights and connections for efficient neural networks. NeurIPS.

    [4] Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., & Chen, L. C. (2018). MobileNetV2: Inverted residuals and linear bottlenecks. CVPR.

    [5] Howard, A. G., et al. (2017). MobileNets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861.

    [6] Elsken, T., Metzen, J. H., & Hutter, F. (2019). Neural architecture search: A survey. JMLR, 20(55), 1–21.

    [7] Real, E., et al. (2019). Regularized evolution for image classifier architecture search. AAAI.

    [8] Cai, H., Zhu, L., & Han, S. (2019). ProxylessNAS: Direct neural architecture search on target task and hardware. ICLR.

    [9] Tan, M., Chen, B., Pang, R., Vasudevan, V., Sandler, M., Howard, A., & Le, Q. V. (2019). MnasNet: Platform-aware neural architecture search for mobile. CVPR.

    [10] Yang, T. J., Howard, A., Chen, B., Zhang, X., Go, A., Sze, V., & Adam, H. (2018). NetAdapt: Platform-aware neural network adaptation for mobile applications. ECCV.

    [11] Lin, J., Rao, Y., Lu, J., & Zhou, J. (2020). Neural architecture search for lightweight non-local networks. CVPR.

    [12] Wu, B., Dai, X., Zhang, P., Wang, Y., Sun, F., Wu, Y., & Keutzer, K. (2019). FBNet: Hardware-aware efficient convnet design via differentiable neural architecture search. CVPR.

    [13] Lane, N. D., Bhattacharya, S., Mathur, A., Georgiev, P., Forlivesi, C., Kawsar, F., & Seneviratne, A. (2016). DeepX: A software accelerator for low-power deep learning inference on mobile devices. IPSN.

    [14] Sze, V., Chen, Y. H., Yang, T. J., & Emer, J. S. (2017). Efficient processing of deep neural networks: A tutorial and survey. Proceedings of the IEEE, 105(12), 2295-2329.

    [15] Zhang, X., Zhou, X., Lin, M., & Sun, J. (2018). ShuffleNet: An extremely efficient convolutional neural network for mobile devices. CVPR.

  • Downloads