臺灣博碩士論文加值系統

意識到那些行動影像的潛在價值，基於邊緣和雲端計算的的一個 行動影像收集平台被推出，嘗試去更有效地利用這些影像。然而，因 為隱私權疑慮跟頻寬的要求，這個平台無法將所有的影像傳回雲端， 基於這個原因，我們想要更完善的去利用這些被留在前端的資料來增 進模型的表現，因此，在這篇論文中，我們為了邊緣計算提出了一個 分散式半監督學習的架構，這個系統不只可以避免使用者上傳他們一 些較敏感的資料，同時也減輕頻寬的需求，我們也使用了現實中的資 料來評估我們的系統，並探討了模型的表現基於一些前端硬體上的限 制。

Recognizing the potential data value of the videos generated from ubiqui- tous personal devices (e.g., dash cams or smartphones), a video collection and analysis platform based on edge/fog and cloud computing is proposed to col- lect and utilize those videos effectively. However, for such a platform, due to the privacy and bandwidth issue, not all of the videos can transmit back to the cloud. We want to fully utilize these left data to increase the model performance further. Thus, in this thesis, we propose a novel distributed ar- chitecture for edge learning, which adopts semi-supervised techniques. The proposed system not only prevents from uploading the sensitive data but also reduce the communication cost. We then evaluate the performance of the sys- tem using real-world video data with a discussion on the performance impact of the hardware limitation at the edge.

口試委員會審定書 i
致謝 ii
中文摘要 iii
Abstract iv
Contents v
List of Figures vii
List of Tables viii
1 Introduction 1
1.1 Background and motivation .................. 1
1.2 Relatedwork ................................ 2
1.3 Contribution................................. 3
2 Scenario and System Model 4
2.1 Scenario................................... 4
2.2 SystemModel ............................... 5
3 Methodology Design 6
3.1 OmniLabel ................................. 6
4 Performance Evaluation 8
4.1 Datasets and model setup.................................8
4.2 Experiment on City Bus.................................. 9
4.3 ExperientonCIFAR-10................................. 10
5 Conclusion 12
Bibliography 13

[1] WeiLiu,DragomirAnguelov,DumitruErhan,ChristianSzegedy,ScottReed,Cheng- Yang Fu, and Alexander C Berg. Ssd: Single shot multibox detector. In European conference on computer vision, pages 21–37, 2016.
[2] Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient con- volutional neural networks for mobile vision applications. arXiv preprint arXiv: 1704.04861, 2017.
[3] H Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, et al. Communication-efficient learning of deep networks from decentralized data. arXiv preprint arXiv:1602.05629, 2016.
[4] Jeffrey Dean, Greg Corrado, Rajat Monga, Kai Chen, Matthieu Devin, Mark Mao, Andrew Senior, Paul Tucker, Ke Yang, Quoc V Le, et al. Large scale distributed deep networks. In Advances in neural information processing systems, pages 1223–1231, 2012.
[5] Jianmin Chen, Xinghao Pan, Rajat Monga, Samy Bengio, and Rafal Jozefowicz. Re- visiting distributed synchronous sgd. arXiv preprint arXiv:1604.00981, 2016.
[6] RyanMcDonald,KeithHall,andGideonMann.Distributedtrainingstrategiesforthe structured perceptron. In Human Language Technologies: The Annual Conference of the North American Chapter of the Association for Computational Linguistics, pages 456–464, 2010.
[7] Sixin Zhang, Anna E Choromanska, and Yann LeCun. Deep learning with elastic averaging sgd. In Advances in Neural Information Processing Systems, pages 685– 693, 2015.
[8] Reza Shokri and Vitaly Shmatikov. Privacy-preserving deep learning. In Proceed- ings of the 22nd ACM SIGSAC conference on computer and communications security, pages 1310–1321, 2015.
[9] JakubKonečnỳ,HBrendanMcMahan,FelixXYu,PeterRichtárik,AnandaTheertha Suresh, and Dave Bacon. Federated learning: Strategies for improving communica- tion efficiency. arXiv preprint arXiv:1610.05492, 2016.