物聯網時代的來臨，數以萬計的聯網攝影裝置將會產生極大量的資料，使得透過邊緣運算來降低雲端運算瓶頸的需求逐漸增加。其中，隨著深度學習模型與辨識技術之精進，透過結合邊緣計算的大量攝影機（本研究稱為邊緣攝影機）運行大範圍的視訊監控已漸漸廣泛應用在各個智慧生活層面。然而，目前的邊緣攝影機無法順利地使用深度學習模型，且其在C2M (customer-to-machine) 物聯網即服務 (IoT-as-a-Service) 尚未存在對應之客製與共享化商業模式。在模型縮小化的議題上，本研究首先設計了「多階段模型壓縮技術（Multi-stage Model Compression）」，其結合了兩種參數壓縮與架構壓縮的主流壓縮方法，建構出多重壓縮組合機制，使不同壓縮技術能夠截長補短，讓深度學習模型在盡可能兼具準確度的情形下減少其模型大小。另外在模型商業化的議題上，本研究設計了「基於區塊鏈之模型商業化服務」，其透過區塊鏈與智慧合約，針對客戶的不同需求，採動態任務驅動 (dynamic task-driven) 的方式將深度模型佈署至所選定的邊緣攝影機中，除了可重複共用的深度模型外，目標是可依據不同的客戶所感興趣的內容進行個別的事件驅動通知，以達成跨使用者間共享模型之客製化商業模式。在系統的驗證上，目前透過此多階段模型壓縮技術可以使深度學習模型在準確度僅減少5個百分點的情形下，將模型大小減少為原先的五十五分之一，其效能也優於既有的文獻效能。另外，透過基於區塊鏈之模型商業化服務，我們設計了一廣域監控之目標協尋雛形系統來驗證可行性，使用者可透過網頁介面填寫任務請求，系統便解析任務內容來匹配並找出系統儲存庫中適合的深度學習模型，接著透過以太坊 (Ethereum) 的加密貨幣向客戶收取費用後，系統將進一步自動產生以太坊智慧合約並在合乎任務需求的邊緣攝影機上進行動態任務模型佈署。若其他客戶提出相似目標的協尋服務請求時，系統可以判別既有任務模型是否可以共享。即使模型共享，但系統仍可依據不同的任務協尋目標即時提供對應的事件與情境資訊通知。

For the upcoming IoT era, a plethora of data will be generated from IoT-enabled devices. If all of data completely rely on traditional cloud servers to process, the cloud servers and the bandwidth will eventually be overwhelmed. To reduce computing bottlenecks on cloud servers, the demand for edge computing is quickly increasing. With the advances in deep neural network (DNN), wide-area monitoring through a large number of edge cameras has been extensively used in various smart-living scenarios. However, most existing edge cameras cannot successfully run a DNN model due to their inadequate memory sizes for a DNN model. Even if a DNN model can be run, there is no corresponding solution of customizable and sharable business model to leverage DNN models as C2M IoT-as-a-Service (customer-to-machine Internet of Things as a Service). To address above two issues, this study first proposes multi-stage model compression, which combines both parameter and structure compression. The proposed method has multiple-compression stages to minimize model size while maintaining as much accuracy as possible. To commercialize the compressed DNN models, our study further proposed BC (blockchain)-enabled model commercialization which features dynamic task-driven and sharable model deployment on edge cameras. The whole system was built upon a blockchain using smart contracts to meet the needs of various customers for wide-area monitoring applications. Even with sharable DNN models, each customer can obtain custmizable event-driven notifications. In the evaluation, through the multi-stage model compression, the size of a DNN model can be reduced by 55 times at the cost of only 5% decrease in the accuracy, which outperforms the state-of-the-art literature. In addition, through the blockchain-based model commercialization, its prototype for wide area video-surveillance was implemented to verify feasibility. Through a friendly web-enabled interface, a user can fill in requests, which will be parsed to match an appropriate DNN model in the model repository. After charging the customer through the cryptocurrency of Ethereum, the system will automatically generate an Ethereum smart contract and dynamically deploy The associated DNN models for the task on their corresponding edge cameras. Other customers applying for a similar request can share the deployed models. Even with the sharable the models, the system can provide customizable event-driven notifications for each customer.

[1] M. Satyanarayanan, P. Simoens, Y. Xiao, P. Pillai, Z. Chen, K. Ha, et al., "Edge analytics in the internet of things," IEEE Pervasive Computing, pp. 24-31, 2015.
[2] G. Ananthanarayanan, P. Bahl, P. Bodík, K. Chintalapudi, M. Philipose, L. Ravindranath, et al., "Real-Time Video Analytics: The Killer App for Edge Computing," Computer, vol. 50, pp. 58-67, 2017.
[3] A. Banafa, "Eight Trends of the Internet of Things in 2018," IEEE Internet of Things. Jan-2018 [Online]. availabel : https://iot.ieee.org/newsletter/january-2018/eight-trends-of-the-internet-of-things-in-2018, 2018.
[4] Y. Gong, L. Liu, M. Yang, and L. Bourdev, "Compressing deep convolutional networks using vector quantization," arXiv preprint arXiv:1412.6115, 2014.
[5] J. A. Hartigan and M. A. Wong, "Algorithm AS 136: A k-means clustering algorithm," Journal of the Royal Statistical Society. Series C (Applied Statistics), vol. 28, pp. 100-108, 1979.
[6] M. Courbariaux, Y. Bengio, and J.-P. David, "Binaryconnect: Training deep neural networks with binary weights during propagations," in Advances in Neural Information Processing Systems, 2015, pp. 3123-3131.
[7] I. Hubara, M. Courbariaux, D. Soudry, R. El-Yaniv, and Y. Bengio, "Binarized neural networks," in Advances in neural information processing systems, 2016, pp. 4107-4115.
[8] C. Buciluǎ, R. Caruana, and A. Niculescu-Mizil, "Model compression," in Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, 2006, pp. 535-541.
[9] G. Hinton, O. Vinyals, and J. Dean, "Distilling the knowledge in a neural network," arXiv preprint arXiv:1503.02531, 2015.
[10] S. Han, J. Pool, J. Tran, and W. Dally, "Learning both weights and connections for efficient neural network," in Advances in Neural Information Processing Systems, 2015, pp. 1135-1143.
[11] C. A. Walsh, "Peter Huttenlocher (1931–2013)," ed: Nature Publishing Group, 2013.
[12] S. Yao, Y. Zhao, A. Zhang, L. Su, and T. Abdelzaher, "Deepiot: Compressing deep neural network structures for sensing systems with a compressor-critic framework," in Proceedings of the 15th ACM Conference on Embedded Network Sensor Systems, 2017, pp. 1-14.
[13] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, "Dropout: a simple way to prevent neural networks from overfitting," The Journal of Machine Learning Research, vol. 15, pp. 1929-1958, 2014.
[14] S. Nakamoto, "Bitcoin: A peer-to-peer electronic cash system," 2008.
[15] K. Christidis and M. Devetsikiotis, "Blockchains and smart contracts for the internet of things," Ieee Access, vol. 4, pp. 2292-2303, 2016.
[16] M. Conoscenti, A. Vetro, and J. C. De Martin, "Blockchain for the Internet of Things: A systematic literature review," in Computer Systems and Applications (AICCSA), 2016 IEEE/ACS 13th International Conference of, 2016, pp. 1-6.
[17] S. Ibba, A. Pinna, M. Seu, and F. E. Pani, "CitySense: blockchain-oriented smart cities," in Proceedings of the XP2017 Scientific Workshops, 2017, pp. 1-5.
[18] G. Prisco, "Slock. it to Introduce Smart Locks Linked to Smart Ethereum Contracts, Decentralize the Sharing Economy," Bitcoin Magazine. Nov-2015 [Online]. available: https://bitcoinmagazine. com/articles/sloc-it-to-introduce-smart-loc s-lin ed-to-smart-ethereum-contractsdecentralize-the-sharing-economy-1446746719.[Accessed: 20-May-2016], 2016.
[19] Slock.it, "BlockCharge – EV Charging via the. Ethereum BlockChain(HQ)," [Online]. availabel : https://slock.it/solutions.html, 2016.
[20] Grid+, "Grid+ Welcome to the Future of Energy," [Online]. availabel : https://gridplus.io/, 2018.
[21] A. Dorri, M. Steger, S. S. Kanhere, and R. Jurdak, "Blockchain: A distributed solution to automotive security and privacy," IEEE Communications Magazine, vol. 55, pp. 119-125, 2017.
[22] X. Huang, C. Xu, P. Wang, and H. Liu, "LNSC: A Security Model for Electric Vehicle and Charging Pile Management Based on Blockchain Ecosystem," IEEE Access, vol. 6, pp. 13565-13574, 2018.
[23] M. A. Ferrag, M. Derdour, M. Mukherjee, A. Derhab, L. Maglaras, and H. Janicke, "Blockchain Technologies for the Internet of Things: Research Issues and Challenges," arXiv preprint arXiv:1806.09099, 2018.
[24] A. Kapitonov, S. Lonshakov, A. Krupenkin, and I. Berman, "Blockchain-based protocol of autonomous business activity for multi-agent systems consisting of UAVs," in 2017 Workshop on Research, Education and Development of Unmanned Aerial Systems (RED-UAS), 2017, pp. 84-89.
[25] Y. Cheng, D. Wang, P. Zhou, and T. Zhang, "Model Compression and Acceleration for Deep Neural Networks: The Principles, Progress, and Challenges," IEEE Signal Processing Magazine, vol. 35, pp. 126-136, 2018.
[26] S. Ioffe and C. Szegedy, "Batch normalization: Accelerating deep network training by reducing internal covariate shift," arXiv preprint arXiv:1502.03167, 2015.
[27] F. Li and B. Liu, "Ternary Weight Networks," arXiv preprint arXiv:1605.04711, 2016.
[28] Y. Bengio, N. Léonard, and A. Courville, "Estimating or propagating gradients through stochastic neurons for conditional computation," arXiv preprint arXiv:1308.3432, 2013.
[29] A. Bogner, M. Chanson, and A. Meeuw, "A Decentralised Sharing App running a Smart Contract on the Ethereum Blockchain," presented at the Proceedings of the 6th International Conference on the Internet of Things, Stuttgart, Germany, 2016.
[30] J. Benet, "IPFS-content addressed, versioned, P2P file system," arXiv preprint arXiv:1407.3561, 2014.
[31] D. Crockford, "The application/json media type for javascript object notation (json)," 2006.
[32] H. Gupta, A. Vahid Dastjerdi, S. K. Ghosh, and R. Buyya, "iFogSim: A toolkit for modeling and simulation of resource management techniques in the Internet of Things, Edge and Fog computing environments," Software: Practice and Experience, vol. 47, pp. 1275-1296, 2017.
[33] N. Sklavos and O. Koufopavlou, "On the hardware implementations of the SHA-2 (256, 384, 512) hash functions," in Proceedings of the 2003 International Symposium on Circuits and Systems, 2003. ISCAS '03., 2003, pp. V-V.
[34] K. Birman and T. Joseph, Exploiting virtual synchrony in distributed systems vol. 21: ACM, 1987.
[35] H. Delfs and H. Knebl, "Symmetric-key encryption," in Introduction to Cryptography, ed: Springer, 2007, pp. 11-31.
[36] A. Paszke, S. Gross, and A. Lerer, "Automatic differentiation in PyTorch," in Workhop in neural information processing systems, 2017.
[37] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, vol. 86, pp. 2278-2324, 1998.
[38] A. Krizhevsky and G. Hinton, "Convolutional deep belief networks on cifar-10," Unpublished manuscript, vol. 40, 2010.
[39] M. W. Gardner and S. Dorling, "Artificial neural networks (the multilayer perceptron)—a review of applications in the atmospheric sciences," Atmospheric environment, vol. 32, pp. 2627-2636, 1998.
[40] K. Simonyan and A. Zisserman, "Very deep convolutional networks for large-scale image recognition," arXiv preprint arXiv:1409.1556, 2014.
[41] K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770-778.
[42] J. Hu, L. Shen, and G. Sun, "Squeeze-and-excitation networks," arXiv preprint arXiv:1709.01507, vol. 7, 2017.
[43] L. Bottou, F. E. Curtis, and J. Nocedal, "Optimization methods for large-scale machine learning," arXiv preprint arXiv:1606.04838, 2016.
[44] D. P. Kingma and J. Ba, "Adam: A method for stochastic optimization," arXiv preprint arXiv:1412.6980, 2014.
[45] J. Redmon and A. Farhadi, "YOLO9000: Better, Faster, Stronger," in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 6517-6525.