2019-01-012024-05-17https://scholars.lib.ntu.edu.tw/handle/123456789/677814摘要：過去物聯網發展多集中在平臺建置與基本應用，應開始尋求更多創新應用與服務 ，朝向智慧機器世代發展，其關鍵技術包含感測與辨識、機器學習等。在諸多物聯 網的應用中，因此，情感運算(Affective Computing)已經為IBM、MIT Media Lab等 研究中心的重點研究方向，為此，本計畫將開發「多模式情感運算應用之感應器與 系統」之相關關鍵技術。 為了降低物聯網中資料傳輸流量、提升後端資訊分析運算的效率，Cisco提出階層 性「霧運算(Fog Computing)」的概念。但是相對於雲端運算，在階層運算結構下 ，接近感知層的節點上所擁有的記憶體容量、硬體運算資源皆非常有限。另一方面 ，感測器的佈署從多個電極的精密量測轉換到穿載式裝置的簡易量測，造成訊號可 信度與解析度下降，進而增加辨識的困難。 為了解決上述挑戰，克服運算資源有 限之瓶頸與整合各種感測器資料是關鍵，本計畫致力於智慧型物聯網系統關鍵技術 開發，將著重於兩個主軸方向來發展：1) 前端各種智慧式資料擷取設備 (Intelligent Sensing)，以及2) 後端的智慧式資訊分析運算(Intelligent Computing)。 本計畫之研究目標與合作之企業公司-「原相科技」之未來主要研發息息相關，目 前在相關工作項目已經進行兩年產學合作計畫 (2017/4~2019/3)，本期將以第一期 合作研發技術為基礎，繼續進行第二期研發計畫，以期持續深根雙方認可之技術 ，以co-development的心態，定義三項研究主題，共同研發產業所需之關鍵前瞻 技術。 本研究計畫提出三項主題進行研究：1)多模式情感運算學習電路與系統：本計畫將 探討機器學習之硬體實現議題以及如何提升分析準確度。因此，如何藉由多模式學 習有效利用各種感測訊號以達到有效學習結果，並且考量在有限的能源與硬體計算 資源下，實現情感運算演算法；情感運算之重點亦將著重於思緒漫遊以及專注度模 型，形成本主題之研究主軸。2)幾何資料處理平台設計：以一階原始-對偶 方法為 核心，發展立體幾何資料處理平台，除了使用GPU進行計算，也會設計相對應之 硬體處理架構，並進行晶片設計與FPGA驗證。3)智慧型物聯網之感應器設計： 本 科技部計畫研究項目將以石墨烯感測元件為主軸，藉由電漿對石墨烯進行表面改質 ，增進材料感測效果。並搭配感測元件設計，使其成為一符合物聯網需求之低功耗 之氣體感測裝置。<br> Abstract: The development of the Internet of Things (IoT) is concentrated in the platform construction and basic application. We should start to seek more innovative applications and services, and toward the development of intelligent machine generation. The key technologies include sensing, identification, machine learning and so on. Among the various applications in IoT, Affective Computing is an important research topic of IBM and MIT media lab. Therefore, this project expects to achieve a “Multi-modal Affective Computing framework” based on “Imaging and Environmental Sensors.” Cisco had proposed “Fog Computing” in order to reduce the amount of data transmission and increase the computation efficiency of data analysis. However, the computational resources of memory and hardware performance in fog node, which is near sensory layer, are more restricted than the resources in cloud layer. On the other hand, the displacement of sensors has transformed from precise and location-specified measurement to light and wearable device. To solve the aforementioned problems, limitation of computational resource and difficulty of aggregating various imprecise sensors are required to be overcome. To solve the above problems, this project is committed to develop “Enabling Technologies for Intelligent IoT System.” Our team will focus on two main directions: 1) Design of high energy efficient and intelligent Sensing hardware modules, and 2) Development of the light-weight machine learning system on data analysis layers. The objectives of this project are closely related to the major research and development themes of the partner company - “Pixart Imaging Inc.” Therefore, the company had worked with NTU team for a two-year cooperation program with related project items (2017/4~2019/3). Based on the foundation of the 1st-phase collaboration project, both sides propose this 2nd-phase collaboration project. The three research and development themes are aligned with the key research topics of the company-interested technologies. This project proposed three research topics: 1) Multi-modal affective computing circuits and systems design: This project will explore the hardware implementation of machine learning and accuracy improvement. Therefore, we will focus on how to effectively use variety of sensing signals to achieve effective learning results by multi-modal learning, and to implement traditional high- complexity information analysis algorithms in limited energy and hardware computing resources. Finally, we will achieve multi-modal affective computing engine for systematic validation. Also, in this phase, we will also put emphasis on “mind wondering” and “attention model” in the area of Affective Computing. 2) Platform design for 3D geometry processing: Our computing platform for 3D geometry processing is based on First-Order Primal-Dual method. In addition to the GPU version, we will also implement hardware accelerators for geometry processing on ASIC/FPGA using the architecture proposed. 3) Novel sensor design for Intelligent IoT: In this project, graphene gas sensors are the cores we are going to build with. Based-on material modification by plasma treatment and design of new sensing device, we aim to improve the performance of graphene gas sensor in order to meet the requirement of internet-of-things. In the long run, we expect the new sensor to be a powerful total solution for the “device layer” of IoT.多模式情感運算應用之感應器與系統設計(第二期)Multi-modal Affective Computing Based on Imaging and Environmental Sensors (Phase II)