DISTRIBUTION OF THE NEURAL NETWORK BETWEEN MOBILE DEVICE AND CLOUD INFRASTRUCTURE SERVICES
Abstract
Neural networks become the only way to solve problems in some areas. Such tasks as recognition of images, sounds, classification require serious processor power and memory for training and functioning of the network. Modern mobile devices have quite good characteristics for primary layers of deep neural networks, but there are not enough resources for whole network. Since neural networks for mobile devices are trained separately on external resources, a method of distributed work of a neural network with vertical distribution over sets of layers with synchronization of training data was developed. The model is divided after saving its state, all layers on the mobile device are converted to the format for the mobile framework and synchronized with the device after training on a distributed platform. Variables and coefficients are formed separately, which allows to significantly reduce the size of the neural network data file uploaded to the device. An algorithm for automatic selection of a neural network separation point was proposed. It based on the data amount transferred between the layers and the load on the mobile device resources. The approach allows to use full-size deep neural networks with a mobile device. Performance experiment showed possibility of obtains an acceptable response even with an unstable communication channel without overloading communication channels and device resources.
References
[2] Iandola F.N., Han S., W. Moskewicz M.W. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size. arXiv:1602.07360v4 [cs.CV], 2016. 13 p. Available at: https://arxiv.org/pdf/1602.07360.pdf (accessed 23.08.2018).
[3] Mengwei Xu, Mengze Zhu, Yunxin Liu, Felix Xiaozhu Lin, Xuanzhe Liu. DeepCache: Principled Cache for Mobile Deep Vision. Proceedings of the 24th Annual International Conference on Mobile Computing and Networking (MobiCom '18). ACM, New York, NY, USA, 2018, pp. 129-144. DOI: 10.1145/3241539.3241563
[4] Shankar S., Robertson D., Ioannou Y., Criminisi A., Cipolla R. Refining Architectures of Deep Convolutional Neural Networks. Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, 2016, pp. 2212-2220. DOI: 10.1109/CVPR.2016.243
[5] Teerapittayanon S., McDanel B., Kung H.T. Distributed Deep Neural Networks Over the Cloud, the Edge and End Devices. Proceedings of 2017 IEEE 37th International Conference on Distributed Computing Systems (ICDCS). Atlanta, GA, 2017, pp. 328-339. DOI: 10.1109/ICDCS.2017.226
[6] LiKamWa R., Hou Y., Gao Y., Polansky M., Zhong L. RedEye: Analog ConvNet Image Sensor Architecture for Continuous Mobile Vision. Proceedings of 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA). Seoul, 2016, pp. 255-266. DOI: 10.1109/ISCA.2016.31
[7] Loc H.N., Lee Y., Balan R.K. DeepMon: Mobile GPU-based Deep Learning Framework for Continuous Vision Applications. Proceedings of the 15th Annual International Conference on Mobile Systems, Applications, and Services (MobiSys '17). ACM, New York, NY, USA, 2017, pp. 82-95. DOI: 10.1145/3081333.3081360
[8] Mathur A., Lane N.D., Bhattacharya S., Boran A., Forlivesi C., Kawsar F. DeepEye: Resource Efficient Local Execution of Multiple Deep Vision Models using Wearable Commodity Hardware. Proceedings of the 15th Annual International Conference on Mobile Systems, Applications, and Services (MobiSys '17). ACM, New York, NY, USA, 2017, pp. 68-81. DOI: 10.1145/3081333.3081359
[9] Jouppi N. Google supercharges machine learning tasks with TPU custom chip. Google Cloud, 2016. Available at: https://cloud.google.com/blog/products/gcp/google-supercharges-machine-learning-tasks-with-custom-chip (accessed 23.08.2018).
[10] Lovejoy B. Apple moves to third-generation Siri back-end, built on open-source Mesos platform. 9to5mac. 2015. Available at: https://9to5mac.com/2015/04/27/siri-backend-mesos/ (accessed 23.08.2018).
[11] Hauswald J. et al. DjiNN and Tonic: DNN as a service and its implications for future warehouse scale computers. Proceedings of 2015 ACM/IEEE 42nd Annual International Symposium on Computer Architecture (ISCA). Portland, OR, 2015, pp. 27-40. DOI: 10.1145/2749469.2749472
[12] Zhang Q., Yang L.T., Chen Z. Privacy Preserving Deep Computation Model on Cloud for Big Data Feature Learning. IEEE Transactions on Computers. 2016; 65(5):1351-1362. DOI: 10.1109/TC.2015.2470255
[13] Kang Y., Hauswald J., Gao C., Rovinski A., Mudge T., Mars J., Tang L. Neurosurgeon: Collaborative Intelligence Between the Cloud and Mobile Edge. ACM SIGARCH Computer Architecture News. 2017; 45(1):615-629. DOI: 10.1145/3093337.3037698
[14] Zhang Y., Huang G., Liu X., Zhang W., Mei H., Yang S. Refactoring android Java code for on-demand computation offloading. Proceedings of the ACM international conference on Object oriented programming systems languages and applications (OOPSLA '12). ACM, New York, NY, USA, 2012, pp. 233-248. DOI: 10.1145/2384616.2384634
[15] Wang X., Liu X., Zhang Y., Huang G. Migration and execution of JavaScript applications between mobile devices and cloud. Proceedings of the 3rd annual conference on Systems, programming, and applications: software for humanity (SPLASH '12). ACM, New York, NY, USA, 2012, pp. 83-84. DOI: 10.1145/2384716.2384750
[16] Zhang Y., Huang G., Zhang W., Liu X., Mei H. Towards module-based automatic partitioning of java applications. Frontiers of Computer Science. 2012; 6(6):725-740. DOI: 10.1007/s11704-012-2220-x
[17] Full imagenet dataset. GitHub. 2017. Available at: https://github.com/tornadomeet/ResNet/blob/master/README.md#imagenet-11k (accessed 23.08.2018).
This work is licensed under a Creative Commons Attribution 4.0 International License.
Publication policy of the journal is based on traditional ethical principles of the Russian scientific periodicals and is built in terms of ethical norms of editors and publishers work stated in Code of Conduct and Best Practice Guidelines for Journal Editors and Code of Conduct for Journal Publishers, developed by the Committee on Publication Ethics (COPE). In the course of publishing editorial board of the journal is led by international rules for copyright protection, statutory regulations of the Russian Federation as well as international standards of publishing.
Authors publishing articles in this journal agree to the following: They retain copyright and grant the journal right of first publication of the work, which is automatically licensed under the Creative Commons Attribution License (CC BY license). Users can use, reuse and build upon the material published in this journal provided that such uses are fully attributed.
