Uma Arquitetura de Rede Neural com Auxílio da Nuvem para Dispositivos Computacionalmente Limitados

• Caio Gevegir Miguel Medeiros UFRJ
• Pedro Cruz UFRJ
• Rodrigo de Souza Couto UFRJ

DOI: https://doi.org/10.5753/wgrs.2023.741

Abstract

Deep Neural Networks (DNNs) may be unfeasible on constrained devices due to their computational complexity. This work proposes an optimized neural network model that runs on these devices at the cost of lower accuracy, but that is supported by a DNN hosted on the cloud if local inferences do not reach a defined confidence threshold. Experiments using handwritten numerical digits show that the proposed model has low memory usage and lower training and inference times compared to a known DNN. In addition, the strategy of inferring data locally before querying to the cloud reduced the average inference time to at least a quarter of its original time, holding an accuracy of 96%.

References

Aleksander, I., De Gregorio, M., França, F. M. G., Lima, P. M. V. e Morton, H. (2009). A brief introduction to weightless neural systems. Em European Symposium on Artificial Neural Networks Advances in Computational Intelligence and Learning (ESANN), p. 299–305. Citeseer.

Aleksander, I., Thomas, W. e Bowden, P. (1984). Wisard· a radical step forward in image recognition. Sensor review, 4(3):120–124.

Bloom, B. H. (1970). Space/time trade-offs in hash coding with allowable errors. Communications of the ACM, 13:422–426.

Bochie, K., Gilbert, M. S., Gantert, L., Barbosa, M. S. M., Medeiros, D. S. V. e Campista, M. E. M. (2021). A survey on deep learning for challenged networks: Applications and trends. Journal of Network and Computer Applications, 194:103213.

Chen, X., Zhang, J., Lin, B., Chen, Z., Wolter, K. e Min, G. (2021). Energy-efficient offloading for dnn-based smart iot systems in cloud-edge environments. IEEE Transactions on Parallel and Distributed Systems, 33(3):683–697.

Grieco, B. P., Lima, P. M., De Gregorio, M. e França, F. M. (2010). Producing pattern examples from “mental” images. Neurocomputing, 73(7):1057–1064.

Hubara, I., Courbariaux, M., Soudry, D., El-Yaniv, R. e Bengio, Y. (2016). Binarized neural networks. Advances in Neural Information Processing Systems, 29.

Hubara, I., Courbariaux, M., Soudry, D., El-Yaniv, R. e Bengio, Y. (2017). Quantized neural networks: Training neural networks with low precision weights and activations. The Journal of Machine Learning Research, 18(1):6869–6898.

Jacob, B., Kligys, S., Chen, B., Zhu, M., Tang, M., Howard, A., Adam, H. e Kalenichenko, D. (2018). Quantization and training of neural networks for efficient integer-arithmetic-only inference. Em Proceedings of the IEEE Conference on Computer Vision and Patter Recognition, p. 2704–2713.

LeCun, Y., Jackel, L. D., Bottou, L., Cortes, C., Denker, J. S., Drucker, H., Guyon, I., Muller, U. A., Sackinger, E., Simard, P. et al. (1995). Learning algorithms for classification: A comparison on handwritten digit recognition. Neural Networks: The Statistical Mechanics Perspective, 261(276):2.

Matsubara, Y., Levorato, M. e Restuccia, F. (2022). Split computing and early exiting for deep learning applications: Survey and research challenges. ACM Computing Surveys, 55(5):1–30.

McDanel, B., Teerapittayanon, S. e Kung, H. T. (2017). Embedded binarized neural networks. arXiv preprint arXiv:1709.02260.

Pacheco, R., Couto, R. e Simeone, O. (2021a). Calibration-aided edge inference offloading via adaptive model partitioning of deep neural networks. Em IEEE International Conference on Communications (ICC), p. 1–6.

Pacheco, R., Oliveira, F. R. e Couto, R. (2021b). Early-exit deep neural networks for distorted images: providing an efficient edge offloading. Em IEEE Global Communications Conference (GLOBECOM), p. 1–6.

Rajapakse, V., Karunanayake, I. e Ahmed, N. (2022). Intelligence at the extreme edge: a survey on reformable tinyml. ACM Computing Surveys.

Rastegari, M., Ordonez, V., Redmon, J. e Farhadi, A. (2016). Xnor-net: Imagenet classification using binary convolutional neural networks. CoRR, abs/1603.05279.

Santiago, L., Verona, L., Rangel, F., Firmino, F., Menasché, D. S., Caarls, W., Breternitz Jr, M., Kundu, S., Lima, P. M. e França, F. M. (2020). Weightless neural networks as memory segmented bloom filters. Neurocomputing, 416:292–304.

Susskind, Z., Arora, A., Miranda, I. D. D. S., Villon, L. A. Q., Katopodis, R. F., de Araújo, L. S., Dutra, D. L. C., Lima, P. M. V., Franca, F. M. G., Breternitz Jr, M. et al. (2022). Weightless neural networks for efficient edge inference. arXiv preprint arXiv:2203.01479.

Teerapittayanon, S., McDanel, B. e Kung, H. (2017). Distributed deep neural networks over the cloud, the edge and end devices. Em 2017 IEEE 37th International Conference on Distributed Computing Systems (ICDCS), p. 328–339.

Teerapittayanon, S., McDanel, B. e Kung, H.-T. (2016). Branchynet: Fast inference via early exiting from deep neural networks. Em IEEE International Conference on Pattern Recognition (ICPR), p. 2464–2469.

Torres, V. A., Jaimes, B. R., Ribeiro, E. S., Braga, M. T., Shiguemori, E. H., Velho, H. F., Torres, L. C. e Braga, A. P. (2020). Combined weightless neural network FPGA architecture for deforestation surveillance and visual navigation of uavs. Engineering Applications of Artificial Intelligence, 87:103227.

Xue, M., Wu, H., Peng, G. e Wolter, K. (2021). DDPQN: An efficient dnn offloading strategy in local-edge-cloud collaborative environments. IEEE Transactions on Services Computing, 15(2):640–655.