臺灣博碩士論文加值系統

在本論文中，我們提出了一種名為“MiniDeep”的新型 AI 邊緣設備，它提供了一 個基於雲邊緣架構的獨立深度學習平台。這種 AI-edge 設備為開發人員提供了一個完整 的深度學習開發環境，以設置他們的深度學習生命週期過程，如模型訓練、模型評估、 模型部署、模型推理、新資料收集、數據預處理和訓練數據管理。 MiniDeep 使用 AWS 作為深度學習模型訓練之管理的後台。在邊緣設備中，OpenVino 可以在邊緣進行深度學 習推論 (Inference)加速。為了執行深度學習生命週期工作，MiniDeep 提出了一個小型 深度生命週期（MDLC）系統，該系統由雲端到邊緣由幾個 microservice 所組成.MiniDeep 提供用於訓練數據集管理和模型訓練計劃的 Train job creator（TJC）和 Model deployer（MD）用於模型之打包與管理，並且都是基於 AWS 雲服務。在邊緣設備上， MiniDeep 提供一個 Inference Handler（IH），通過來自 End device 的 RESTful API 請 求/回應來處理深度學習之運算需求，而 Data provider（DP）負責將收集到的新資料同 步至雲端服務和數據集同步 API。本文利用深度學習能力，利用 MiniDeep 平台實現了 AI-QSR（QuickService Restaurant）KIOSK 應用的推薦系統。 AI-QSR 使用 MiniDeep 平台來訓練基於 LSTM 的推薦系統。基於 LSTM 的推薦系統將 KIOSK UI 流程轉換為 Flow sequence，並根據結果來向使用者推薦食物。在本論文的最後，我們展示了我們的實驗 環境與數據，並演示了 AI-QSR 推薦系統的性能分析。

In this thesis, we present a new AI-edge device, called "MiniDeep", which provide a standalone deep learning platform based on cloud-edge architecture. This AI-edge device provides developer a whole deep learning development environment to setups their deep learning life cycle process, such as model training, model evaluation, model deployment, model inference, ground truth collecting, data pre-processing, and training data management. MiniDeep uses AWS as our backend platform of deep learning model tuning management. In the edge device, The OpenVino enables deep learning inference acceleration at the edge. To perform deep learning life cycle job, MiniDeep proposes a mini deep life cycle (MDLC) system which is composed of several microservices from cloud to edge. MiniDeep provides Train Job Creator (TJC) for training dataset management and model training schedule, and Model Packager (MP) is for model package management. All of them based on several AWS cloud service. On the edge device, MiniDeep provides Inference Handler (IH) to handle deep learning inference by hosting RESTful API request/response from end device, and Data Provider (DP) is responsible for ground truth collection and dataset synchronization API for cloud. With the deep learning ability, this thesis uses MiniDeep platform to implement a recommendation system for AI-QSR (Quick Service Restaurant) KIOSK application. AI-QSR uses MiniDeep platform to train an LSTM-based recommendation system. The LSTM-based recommendation system converts KIOSK UI-flow to the flow sequence and performs sequential recommendation with food suggestion. At the end of this thesis, this method shows the experiment environment and shows the performance analysis for AI-QSR recommendation system.

List of content . . . i
List of figure . . . iii
1 Introduction . . . 1
2 Related Works . . . 5
3 Preliminaries . . . 8
3.1 Edge-Cloud architecture for deep learning. . . 8
3.2 Deep learning cloud service overview . . . 11
3.3 Deep learning edge environment overview . . . 12
3.4 Problem Definition. . . 13
3.5 Problem Formulation . . . 15
4 MiniDeep Edge-cloud Platform 18
4.1 MiniDeep system architecture . . . 18
4.2 MiniDeep platform usages design . . . 20
4.2.1 Data Provider . . . 20
4.2.2 Train Job Creator . . . 23
4.2.3 Training procedure details . . . 29
4.2.4 Model Deployer . . . 32
4.2.5 Inference engine handler . . . 34
4.2.6 Inference procedure details . . . 35
5 AI-QSR KIOSK Application Software Architecture . . . 41
5.1 AI-QSR system architecture . . . 41
5.2 LSTM-based recommendation system design for AI-QSR . . . 43
5.3 AI-QSR recommendation system implemented by MiniDeep platform . . . 47
6 Experimental Results . . . 49
6.1 Experimental environment . . . 49
6.2 Performance analysis . . . 52
6.2.1 Purchase hit accuracy . . . 54
6.2.2 Categorical crossentropy . . . 54
6.2.3 Precision. . . 55
6.2.4 Recall . . .57
6.2.5 F1 score . . . 57
6.2.6 Cohen’s kappa coefficient . . . 58
6.2.7 Iterations . . . 59
7 Conclusions . . . 61
8 Acknowledgments . . . 62

[1] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu, ”Edge Computing: Vision and Challenges”,
IEEE Internet of Things Journal, vol. 3, no. 5, pp. 637–646, 2016.
[2] X. Liu, Z. Qin, and Y. Gao, ”Resource Allocation for Edge Computing in IoT Networks
via Reinforcement Learning”, in Proceedings of 2019 IEEE International Conference on
Communications (ICC 2019), Shanghai, China, pp. 1–6, May 2019.
[3] P. Li, Y. Luo, K. Wang, and K. Yang, ”Energy Minimization and Offloading Number
Maximization in Wireless Mobile Edge Computing”, in Proceedings of 2018 IEEE Global
Communications Conference (GLOBECOM 2018), Abu Dhabi, UAE, pp. 1–6, Dec. 2018.
[4] J. Lin, W. Yu, N. Zhang, X. Yang, H. Zhang, and W. Zhao, ”A Survey on Internet of
Things: Architecture, Enabling Technologies, Security and Privacy, and Applications”,
IEEE Internet of Things Journal, vol. 4, no. 5, pp. 1125–1142, 2017.
[5] Q. Wang, S. Guo, Y. Wang, and Y. Yang, ”Incentive Mechanism for Edge Cloud Profit
Maximization in Mobile Edge Computing”, in Proceedings of 2019 IEEE International
Conference on Communications (ICC), pp. 1–6, May 2019.
[6] J. A. Stankovic, ”Research Directions for the Internet of Things”, IEEE Internet of Things
Journal, vol. 1, no. 1, pp. 3–9, 2014.
[7] W. Yu, F. Liang, X. He, W. Hatcher, C. Lu, J. Lin, and X. Yang, ”A Survey on the Edge
Computing for the Internet of Things”, IEEE Access, vol. 6, pp. 6900–6919, Nov. 2018.
[8] F. Samie, L. Bauer, and J. Henkel, ”From Cloud Down to Things: An Overview of
Machine Learning in Internet of Things”, IEEE Internet of Things Journal, vol. 6, no. 3,
pp. 4921–4934, 2019.
[9] M. Moradi, K. Sundaresan, E. Chai, S. Rangarajan, and Z. M. Mao, ”SkyCore: Moving
Core to the Edge for Untethered and Reliable UAV-based LTE Networks”, in Proceedings of the 24th Annual International Conference on Mobile Computing and Networking
(MOBICOM 2018), New York, NY, USA, pp. 35–49, Nov. 2018.
[10] N. Mohan, T. Shreedhar, A. Zavodavoski, O. Waltari, J. Kangasharju, and S. K. Kaul,
”Redesigning MPTCP for Edge Clouds”, in Proceedings of the 24th Annual International
Conference on Mobile Computing and Networking (MOBICOM 2018), New York, NY,
USA, pp. 675–677, Nov. 2018.
[11] A. Khan, F. Freitag, V. Vlassov, and P. Ha, ”Demo abstract: Towards IoT service deployments on edge community network microclouds”, in Proceedings of IEEE Conference
on Computer Communications (INFOCOM 2018), Honolulu, HI, USA, pp. 1–2, April
2018.
[12] W. Zhang, S. Li, L. Liu, Z. Jia, Y. Zhang, and D. Raychaudhuri, ”Hetero-Edge: Orchestration of Real-time Vision Applications on Heterogeneous Edge Clouds”, in Proceedings
of IEEE Conference on Computer Communications (INFOCOM 2019), Paris, France,
pp. 1270–1278, April 2019.
[13] D. Wu, F. Wang, X. Cao, and J. Xu, ”Wireless Powered User Cooperative Computation
in Mobile Edge Computing Systems”, in Proceedings of IEEE Global Communications
Conference (GLOBECOM 2018), Abu Dhabi, UAE, pp. 1–7, Dec. 2018.
[14] J. Hochstetler, R. Padidela, Q. Chen, Q. Yang, and S. Fu, ”Embedded Deep Learning
for Vehicular Edge Computing”, in Proceedings of 2018 IEEE/ACM Symposium on Edge
Computing (SEC 2018), Bellevue, WA, pp. 341–343, Oct. 2018.
[15] D. Velasco-Montero, J. Fernndez-Berni, R. Carmona-Galn, and . Rodrguez-Vzquez, ”Optimum Selection of DNN Model and Framework for Edge Inference”, IEEE Access, vol. 6,
pp. 51680–51692, Sep. 2018.
[16] R. Rothe, R. Timofte, and L. Van Gool, ”Deep Expectation of Real and Apparent Age
from a Single Image Without Facial Landmarks”, International Journal of Computer
Vision, vol. 126, no. 2-4, pp. 144-157, 2018.
[17] S. Zagoruyko and N. Komodakis, ”Wide Residual Networks”, arXiv preprint
arXiv:1605.07146, 2016.
[18] J. Tang and K. Wang, ”Personalized Top-N Sequential Recommendation via Convolutional Sequence Embedding”, arXiv preprint arXiv:1809.07426, 2018.
[19] R. He, W. Kang, and J. McAuley, ”Translation-based Recommendation”, arXiv preprint
arXiv:1707.02410, 2017.
[20] B. Hidasi, A. Karatzoglou, L. Baltrunas, and D. Tikk, ”Session-based Recommendations
with Recurrent Neural Networks”, arXiv preprint arXiv:1511.06939, 2016.
[21] M. Quadrana, A. Karatzoglou, B. Hidasi, and P. Cremonesi, ”Personalizing Sessionbased Recommendations with Hierarchical Recurrent Neural Networks”, arXiv preprint
arXiv:1706.04148, 2017.
[22] D. Jannach and M. Ludewig, ”When Recurrent Neural Networks Meet the Neighborhood
for Session-Based Recommendation”, in Proceedings of the Eleventh ACM Conference on
Recommender Systems (RecSys 2017), New York, NY, USA, pp. 306–310, Aug. 2017.
[23] “OpenVino.” https://software.intel.com/en-us/openvino-toolkit.
[24] “yu4u/age-gender-estimation.” https://github.com/yu4u/age-gender-estimation.