電機資訊學院: 電信工程學研究所指導教授: 陳光禎莊子由Chuang, Tzu-YuTzu-YuChuang2017-03-062018-07-052017-03-062018-07-052016http://ntur.lib.ntu.edu.tw//handle/246246/276381在社會科學、工程、財金分析、以及未來希望的物聯網等眾多科技領域中,我們都可以發現網路結構式資料的蹤跡。而如何處理網路結構式資料,已為人類社會活動帶來了莫大的機會,並不斷挑戰資料科學家們的心智。一方面來說,研究網路結構式資料的價值在於;如何有效地開發不同資料群組之間的互動與應用,包括高效率的資料傳輸及資料分享、以及在資料分享同時所帶來的資料安全與隱私問題。另一方面,如同時下最熱門的主題「大數據」一般,網路結構式資料也具備大量的取樣以及極高的維度,為資料科學家在計算及統計上創造新的挑戰。這些機會及挑戰促成了這篇論文的生成,旨在提供讀者一賭現今資料科學在網路結構式資料上的發展及應用。以理論發展為經,工程應用為緯的交織下,我們發現網路結構式資料的分析需要比傳統分析更具高度的思維。除了資料本身具備的意義之外,網路結構所帶來的資訊也對資料分析的結果有著決定性的影響。透過檢視這些影響,我們除了可以提升資料分析的準度及精度外,還可以解決許多工程上的問題。我們將這類處理資料的思維稱為網路結構式資料的認知,並將這種結合統計決策及網路結構資訊的分析方法應用於各種不同的情境與實驗中。在工程方面,透過網路結構式資料的認知,我們得以開發有效的無線感測器的管理。在分析河川揚塵及股票市場資料方面,網路結構式資料的認知能夠建立比傳統方法更為精準及有效的預測模型。這些實驗驗證了我們的想法,並為未來資料科學的研究開展了新的方向。我們期望資料科學能在未來的某一天讓資分析具備真正的「智慧」,而網路結構式資料的認知能在這條路上擔任一塊基石。Networked data can be found in many field, including social science, engineering, financial analysis and the Internet of Things. The processing of network data brings new opportunities to our society and challenges to data scientists. On the one hand, the network structure underlying the data holds great promises for utilizing the interaction among different groups of data sources, including efficient data transmission and data sharing, as well as the challenges of privacy preserving and inference attack. On the other hand, like “Big Data”, the massive sample size and high dimensionality of data introduce unique computational and statistical challenges. These opportunities are distinguished and require new computational and statistical paradigms. This dissertation gives an overview on what is networked data and how networked data impact on paradigm changes of analysis techniques and new data engineering architectures. We also provide various perspectives on the networked data analysis and computation. In particular, we emphasize the recognition on networked data, which is a new philosophy that incorporates higher order network structures to solve decision problems on networked data, and point out that decisions incorporating network structure can greatly improve the performance of systems as well as mitigate several engineering problems, including data recovery, privacy preserving and inference attack. Several applications based on networked data analysis are also introduced, including sensor network management, river dust analysis, and interaction between stock markets and exchange rate.5825641 bytesapplication/pdf論文公開時間: 2016/8/25論文使用權限: 同意有償授權(權利金給回饋學校)網路結構式資料認知Networked DataRecognition網路結構資料之認知Recognition on Networked Datathesis10.6342/NTU201603280http://ntur.lib.ntu.edu.tw/bitstream/246246/276381/1/ntu-105-D99942022-1.pdf