研究領域 Research Topics

研究簡介   Click
Introduction   Click
科技部計畫 Project Lists   Click
產學合作計畫 Project Lists   Click

研究簡介

容錯(fault tolerance)技術之研發一般區分為兩個方向:硬體容錯(hardware fault tolerance)與軟體容錯(software fault tolerance)。傳統的容錯研究幾乎均著重於硬體容錯,關於軟體或系統的容錯研究可謂是一門新的領域。軟體容錯,簡單地說就是包含所有在軟體中廣泛被使用來解決各種不同類型錯誤的方法,一如硬體容錯中使用的各種方法來對付不同種類的錯誤。進一步而言,軟體容錯技術 (Software Implemented Fault Tolerance),所指的是利用一組軟體來偵測作業系統或硬體方面所無法處理的錯誤以及從錯誤的情況加以還原。在此我們所考慮的錯誤是指會造成應用程式的損毀 (crash) 或懸滯 (hang)。有了軟體容錯技術之後,許多工商業界的應用程式便可以最有效率的方式將錯誤發生時的損失減到最小。

雲端運算平台之核心技術即在於虛擬機管理平台,本計畫針對以工業電腦為硬體平台之虛擬機容錯服務(fault-tolerant virtual machines, FT-VM)相關技術進行研究。工業用電腦目前的應用相當廣泛,如:通訊與網路、工業自動化、醫療運算、…等等,皆是主要應用市場。這些應用對於系統可靠性與穩定性都有高度之要求。然而最近,國內廠商發現未來工業電腦產品不再是比硬體品質或價格的市場,而是必須在以符合工業標準例如IMPI的產品再加上各種領域相關(domain specific)之加值應用(value added applications)。展望未來,在IMPI刀鋒伺服器上導入具高可靠度與高容錯性之雲端虛擬機服務是必然的趨勢,而目前這項技術(FTVM over IMPI)還正在起步中。國內廠商目前急需下列技術來提升產品的品質與附加價值:導入軟體容錯技術、研發軟硬體系統之整合與效能之測試技術研發、開發特定工業用途之加值服務等等。

過去數年,我們在Intel x86平台上研究開發軟硬體整合高可靠度與容錯之虛擬機技術,以符合國內廠商產品品質與軟體服務升級的需求。我們持續研發軟硬體整合的產品測試技術,確保我們研發軟體的品質,同時也能提升廠商新產品的品質。在人才培育部分,我們以技術導向實務為目標,進行教案編撰與課程開設。中大團隊在過去一年累積可觀的成果:(1)我們研發的「記憶體層級之不停機容錯(Zero-Down-time Fault Tolerance)技術」,經過測試軟體驗證後,證實其效能勝過目前的產品Micro-Checkpoint:我們的容錯虛擬機技術的每分鐘執行指令是Micro-checkpoint的2.85倍,服務等待時間(response time)比Micro-Checkpoint減少了67%,單位指令頻寬消耗則是Micro-checkpoint的41%。(2) 在軟硬體測試方面,目前國內工業現狀大部分只能以人工測試部分功能,而我們已能提升廠商品質確保能力至全面性的回歸測試,大量減少測試的人力與時間。以合作廠商為例,人力研發成本降低70%,測試時程縮短80%,新產品開發速度加快50%。(3) 在提升產學界無縫接軌方面,我們積極邀請業師上課,共計30人時。(4) 在技術擴散方面,我們積極與國內廠商以及相關法人接觸洽談合作事項,今年合作廠商共投入268萬研發資金、捐贈價值120萬硬體設備、人力資源投入價值270萬、衍生產學合作案140萬元 (另有多案洽談中)、並將與國內廠商及法人機構爭取大型整合性計畫。


Introduction

Fault tolerance is generally classified into two categories: the hardware fault tolerance and the software fault tolerance. The former receives significant attention from the traditional fault tolerance research community; on the contrary, the latter receives less attention and is regarded as a relatively novel field.

Software Implemented fault tolerance deals with faults that cannot be dealt with by hardware fault-tolerant methods via the use of the software-based detection and recovery methods, which include software hangs and crashes. With software implemented fault tolerance, the possibility of business operational loss due to the software malfunction can be minimized.

The foundation of cloud computing platform is the virtual machine management system (VMM). As a result, the fault tolerance of the VMM that is installed on Industrial personal computers (IPCs) becomes increasingly important as the cloud computing applications burgeon across the IT industry today (IPCs have been widely employed as the core components of many applications, such as networking devices, automation control devices, medical electronic equipment... etc). These applications are known for their strict demands on reliability and robustness while operating in extreme working conditions. Our industry partners have found that the cost and quality of the hardware are no longer the key factors to achieve high profit margin, instead, the ability to develop domain-specific and value-added applications over the IPMI platform, a high-end IPC, have become increasingly important in market competitiveness in the future. Consequently, the IPMI-enabled servers with value-added virtualization technology will be the key products for the next generation of IPCs, since they provide high flexibility and yet maintain high availability. However, the technology to integrate high-availability IPCs and virtualization technology is still at its infantile stage. To improve its competitiveness in the next three to five years, the IPC industry in Taiwan will need to develop the following four key technologies: fault-tolerance for virtualization on the IPMI servers, software and hardware integrated testing technology, as well as value-added applications based on IPMI servers. Therefore, we plan to provide fault tolerance technology for virtualization, as well as automatic testing technologies and tools. Our team also plan to develop new industrial-technology-oriented courses, to provide related teaching material, and to train our students accordingly.The NCU team has obtained the following achievements in the past year: (1) the fault tolerant VM is about three times as fast as the existing tool, Micro-Checkpointing, when measured in operations per minute, (2) the automatic regression test technology and tools that can reduce the testing time up to 80% in product line, (3) the offering of two practical courses featuring lecturers from the industry, and (4) the engagement of joint research projects with our industrial partners.


科技部計畫 Project Lists

執行中 On Going Projects
  • 虛擬化高品質軟體服務核心技術之研發(2016/10-2020/9)(PDF)
    On the Development of Core Technologies for High Quality Virtualized Software Services(PDF)
已結案 Closed Projects
  • 工業用核心虛擬技術之開發(2015/1-2015/12)(PDF)
    On the Development of the Virtual Technology for Industrial Computers(PDF)
  • 工業用核心虛擬技術之開發(2014/1-2014/12)(PDF)
    On the Development of the Virtual Technology for Industrial Computers(PDF)
  • 工業用核心虛擬技術之開發(2013/1-2013/12)(PDF)
    On the Development of the Virtual Technology for Industrial Computers(PDF)
產學合作計畫 Project Lists

執行中 On Going Projects
  • 工業用物聯網測試平台之開發(2016/4/1-2018/12/31) —合作對象:凌華科技股份有限公司
  • 高可靠雲端檔案管理測試平台之先期研究(2016/6/1-2017/12/31) —合作對象:以柔資訊股份有限公司
已結案 Closed Projects
  • 服務導向IOT資料擷取平台雛形研究(2011/4/1-2011/12/20) —合作對象:資策會
  • 工業用電腦虛擬核心技術之開發(2013/1/1-2015/12/31) —合作對象:凌華科技股份有限公司
  • 虛擬機器之高可靠性架構研究(2015/3/1-2015/11/30) —合作對象:資策會
  • 自動化測試平台之入口網頁服務開發與測試(2015/8/1-2015/12/31) —合作對象:工業技術研究院
  • 物聯網應用自動化測試技術平台網頁服務(2016/4/15-2016/5/27) —合作對象:工業技術研究院

研究簡介   Click
Introduction   Click
科技部計畫 Project Lists   Click
產學合作計畫 Project Lists   Click

研究簡介

由於資訊技術與網際網路的進步,手持行動裝置(如:智慧型手機、平板電腦)的效能與功能迅速地增加。行動裝置不再只是作為使用者的通訊工具,還可用在商業應用服務上,如行動銀行與線上購物。而這些新應用服務也掀起新的安全議題,如手持行動裝置上的隱私與敏感性資訊如信用卡號碼被洩漏造成財務上的損失。

目前智慧型手機通防護方面,通常是以PIN碼、密碼為主,甚至是採用指紋或臉部等生物特徵的識別技術。然而,密碼與指紋這類型的安全性機制皆需要使用者執行一個特定的認證程序,屬於侵入式驗證的方式。

近幾年,研究人員提出行動裝置上的非侵入式驗證代替侵入式驗證方法,如使用步態、觸碰與姿態模式和手機動作等。雖然非侵入式的驗證方法需要從使用者上擷取少許的行為。但市調指出,有60%-80%用戶會因侵入式認證不方便而選擇關閉該功能。這也是非侵入式驗證會比侵入式驗證更讓使用者接受的原因,此外,非侵入式驗證也因為使用者的行為習慣難被模仿而可以做為更進階的劫持防護機制。


根據2014年的統計報告顯示每天約有3500人因交通事故傷亡。此外,研究也顯示大部分的交通事故是人為因素造成,如駕駛者的異常行為(疲勞、酒駕等等因素),因此,偵測駕駛者異常行為的研究是避免交通意外事故中重要的一環。

近幾年,穿戴型裝置蓬勃發展,相關的產品越來越多元化,其中,智慧型手錶內建多種感測器,如加速度計、陀螺儀與方位感測器等等,可以被用來持續行為監控分析或健康狀況監控,而我們將這些感測器用在收集駕駛者行為資料。

有了駕駛者行為資料之後,根據多位駕駛者資料分布的分析,我們採用高斯混合模型建構每位駕駛者各自的行為模型。高斯混合模型在過去的文獻中,已經被廣泛地應用在不同的領域,如語者識別和駕駛者識別。

目前基於高斯混合模型方法再加以改良提出更準確且強建的系統架構,其研究成果顯示我們的系統架構比傳統的高斯混合模型方法有顯著改善,識別率的度量指標”相等錯誤率”可達4~7%,未來我們團隊也是會朝降低相等錯誤率持續研究下去並且思考如何讓系統有更多的應用。


Introductions

With the advances in information and technology, the performance and features of hand-held devices such as smartphone and tablet are rapidly increased. This enables users to use such devices not only as communication tools but also in business applications such as m-banking and online shopping. These new applications raise concerns about security issues for smartphone users. Personal sensitive data stored in the phone such as credit card number and user credentials should be protected well in order to prevent financial loss when the device is stolen.

The current protection mechanisms of these devices are usually based either on PIN codes, passwords or biometric-based methods, such as fingerprints or iris. Both fingerprints and password entry are intrusive in the sense that they require explicit action from the user. Fingerprints and password based authentication also vulnerable to be compromised when attacker have credentials needed for opening the devices.

Recently, researchers proposed the use of non-intrusive authentication methods using gait, touch and orientation patterns, and phone movement as a replacement for intrusive authentication. The non-intrusive authentication method require minimal explicit action form the user. Market research points out that there are about 60%-80% of smartphone users who choose to turn off intrusive authentication methods due to its inconvenience. That is the reason why non-intrusive authentication methods will more acceptable to the users. The non-intrusive authentication methods also promising advanced hijacking protection due to its difficulty to mimic other person’s behavior.

In 2014, traffic accidents claimed nearly 3500 lives each day. Studies reveal that most traffic accidents are caused by human factors, such as drivers’ abnormal driving behaviors. Therefore, detection of the abnormal driving behavior is one of the significant studies of preventing traffic accidents.

The smartwatch, which is equipped with many sensors, such as the accelerometer and orientation sensor, can be used for not only continuous motion analysis and health monitoring but also analysis of the driving behavior.

The Gaussian mixture model (GMM), which is a mixture of Gaussian component densities, is used to model the distribution of the hand-movement feature of the driver obtained from the smartwatch. In the literature, such as text-independent speaker recognition and driver recognition, GMMs are often used in recognition issues.

In recent, we have proposed a more robust and accurately system architecture based on traditional GMM approach. Then, the research results show our proposed system has significant improvements than the transitional GMM approach, and the equal error rate (EER) of our proposed system can reach 4~7%. In future work, the team members will keep on improving the EER, and try to apply this system to different applications.




科技部計畫 Project Lists

執行中 On Going Projects
  • 支援行動應用APP之惡意風險分析及使用者認證技術研發-總計畫暨子計畫一:智慧型手機可適應多重操作姿勢的多模式使用者驗證機制研發(2016/8)(PDF)
    The Development of Poses Adaptive and Multimodal User Authentication Mechanism for Smartphone Users: The Development of Poses Adaptive and Multimodal User Authentication Mechanism for Smartphone Users(PDF)
已結案 Closed Projects
  • 支援在企業雲環境BYOD應用之資訊安全技術研發-以行為特徵為基礎之可適應多重操作姿勢的行動裝置使用者認證技術研發(2015/8-2016/7)(PDF)
    A poses adaptive authentication mechanism based on behavioral biometrics obtained from mobile devices(PDF)
  • 可支援雲端服務平台之行動使用者認證鎖技術研發(2014/8-2015/7)(PDF)
    A handheld device authentication mechanism for Cloud services(PDF)
  • 支援行動裝置使用者與虛擬實驗平台之雲端技術研究--雲客端-手持行動裝置使用者識別機制(2013/8-2014/7)(PDF)
    A handheld device authentication mechanism to improve security level of Cloud services via securing the Cloud clients(PDF)
  • 用於維護雲端服務(含客端)安全機制與其安全服務實驗平台之研究-總計畫(2010/8-2013/7)
    Developing a Mechanism and an Establishment of Testbed for Cloud Security Services
  • 用於維護雲端服務(含客端)安全機制與其安全服務實驗平台之研究-子計畫一:用於雲端服務之行動裝置與網際網路服務異常行為監控與防護機制之研究(2010/8-2013/7)(PDF)
    A Monitoring and Protecting Mechanism for Abnormal Behavior of Mobile and Web services in Cloud Computing Environment(PDF)

產學合作計畫 Project Lists

執行中 On Going Projects
  • 高可靠雲端檔案管理測試平台之先期研究(2016/6/1-2017/12/31) —合作對象:以柔資訊股份有限公司
已結案 Closed Projects
  • 全周即時主動安全感測技術(2010/6/1-2010/11/30)—合作對象:工業技術研究院

研究簡介   Click
Introduction   Click
科技部計畫 Project Lists   Click

研究簡介

有效預測財務困難的企業對於金融機構作出適當的貸款決策是至關重要的。在一般情況下,輸入的變數(或特徵),如財務指標、特徵篩選方法的選擇、以及使用統計方法和機器學習技術,是影響預測效能的三個重要因素。在過去的十年中,我們的研究團隊在三個方向上對於財務危機預測問題(financial distress prediction problem , 簡稱: FDP)都獲得了巨大的啟發。在最近的研究中,公司治理指標(corporate governance indicators , 簡稱: CGIs)已經被發現是輸入變數中的除了財務指標(financial ratios , 簡稱: FRs)之外另一個重要的特徵。然而,通過合併CGIs和FRs獲得的效能還沒有被完整檢驗,因為只有一些選定的CGIs和FRs在相關的研究中使用,而所選擇的特徵在各個研究中有可能完全不同。因此,我們的研究致力於這個方向,以評估分別結合不同類別的 FRs和CGIs 所獲得的預測效能。實驗結果根據台灣的上市櫃公司真實資料,顯示在FRs中的償債能力和盈利能力和CGIs中的董事會結構和股權結構都是預測破產最重要的特徵。特別是,從預測的準確率、第I / II型錯誤、ROC曲線、以及分類錯誤成本的組合中得到的最佳預測模型的效能。更多詳細信息請詳閱我們最近的著作[LLT16] [LLL15]

許多研究認為在建構模型之前,特徵篩選是為資料探勘的預先處理步驟。大多數的研究主要集中在於應用單一特徵篩選方法,我們進行了全面的研究,探討進行filter approach和wrapper approach特徵篩選方法的效果。同時,特徵篩選對使用多種classifiers預測模型的影響也進行了研究。在實驗中,使用了兩個財務危機的資料集。此外,三個filter approach和兩個wrapper approach的特徵篩選方法結合六個相異的預測模型進行了研究。我們的實驗結果顯示,filter approach特徵篩選方法的效能比wrapper approach的模型更佳。另外,根據所選擇的技術中,執行特徵篩選並不總是能夠提高預測效能。有興趣的讀者可以參考[LTW15] [LLY14]

最後,使用適當的機器學習技術來建構預測模型也是影響其性能的關鍵因素。我們引入了classifier ensemble方法,以減少分類錯誤的成本。通過利用unanimous voting(UV)方法結合由多個classifiers產生的輸出,以找到最終的預測結果。基於四個相關的資料集獲得的實驗結果表明,我們的UV集合方法優於許多基線單一分類和classifier ensemble。更具體而言,UV集成不僅提供相對良好的預測準確率和第I / II型錯誤,也僅產生最小的錯誤分類成本。


Introduction

Effective prediction of financially distressed firms is critical for financial institutions to make appropriate lending decisions. In general, the input variables (or features), such as financial ratios, the choice of feature selection methods, and the use of appropriate statistical and machine learning techniques, are the three important factors that affect the prediction performance. In the past decade, our research team has gained tremendous insights on the financial distress prediction problem (or FDP) in all three directions. In the recent studies, corporate governance indicators (CGIs) have been found to be another important type of input variables in addition to financial ratios (FRs). However, the performance obtained by combining CGIs and FRs has not been fully examined since only some selected CGIs and FRs have been used in related studies and the chosen features may differ from study to study. Therefore, the research aim in this direction is to assess the prediction performance obtained by combining seven and five different categories of FRs and CGIs respectively. The experimental results, based on a real-world dataset from Taiwan, show that the categories of solvency and profitability in FRs and the categories of board structures and ownership structures in CGIs are the most important features in bankruptcy prediction. In particular, the best prediction model performance is obtained with combination of prediction accuracy, Type I/II errors, ROC curve, and the misclassification cost. More details can be found in our recent publications [LLT16][LLL15].

Since disputes remain regarding financial ratios as input features for model development, many studies consider feature selection as a pre-processing step in data mining before constructing the models. Apart from most studies, which have focused on applying one specific feature selection methods to FDP, we have conducted a comprehensive study to examine the effects of performing filter and wrapper based feature selection methods. In addition, the effect of feature selection on the prediction models obtained using various classification techniques is also investigated. In the experiments, two financial distress datasets are used. Moreover, three filter and two wrapper based feature selection methods combined with six different prediction models are studied. Our experimental results indicate that filter based feature selection methods perform better than models that are wrapper based. Moreover, depending on the chosen techniques, performing feature selection does not always improve the prediction performance. Interested readers can refer to [LTW15][LLY14].

Last but not least, the use of machine learning techniques to construct a prediction model is also a key factor to its performance. We introduce a classifier ensemble approach to reduce the misclassification cost. The outputs produced by multiple classifiers are combined by utilizing the unanimous voting (UV) method to find the final prediction result. Experimental results obtained based on four relevant datasets show that our UV ensemble approach outperforms many baseline single classifiers and classifier ensembles. More specifically, the UV ensemble not only provides relatively good prediction accuracy and Type I/II errors, but also produces the smallest misclassification cost.

科技部計畫 Project Lists

執行中 On Going Projects
  • 機器學習導入電子產業智慧生產技術之開發與應用(2016/11-2017/10)(PDF)
    Productivity 4.0 for ICT Industry using Machine Learning Technologies(PDF)