由於電子硬體技術與無線網路的迅速發展，促使智慧行動裝置快速發展，其產品涵蓋智慧型手機、平板電腦、智慧型手錶與手環等，其中，智慧型手機更成為存取資訊的重要裝置。然而，根據市調報告指出，有60%-80%的使用者會關閉裝置上的身分認證機制，而且文字或圖形密碼在開放環境中使用無法避免有心人士肩窺。因此，智慧型手機上的安全性需要提升與強化。根據本團隊最近幾年執行的研究計畫與學術發表的論文，已針對智慧型手機研發基於使用者行為的驗證機制；我們發現驗證機制除了考量驗證準確率之外，應將系統之使用性(Usability)與可靠性(Reliability)也納入考量中，如在建構使用者預測模型階段，不適合讓使用者重複太多次相同的收集動作與花費太多時間在收集資料上，這會造成使用者體驗上覺得枯燥，進而降低使用者的使用意願。此外，驗證機制也必須因應使用者可能隨時間改變操作行為，進而必須自動更新預測模型的問題，否則將造成預測效能降低。故本研究計畫將針對智慧型手機使用者認證機制提出一個基於多種學習策略改善智慧型手機使用者驗證機制之模型建構效率的方法，藉此提升認證機制的使用性與可靠度，提高使用者的使用意願與維持驗證效能。 ;With the rapid advances in information and communication technologies, smart mobile devices, such as tablets, smart phones, and smart watches, are widely used. These smart devices are often used to store sensitive personal data. Recent surveys show that more than 60% of smart device users opt to turn off their user authentication mechanisms due to their intrusive nature in order to increase usability. We therefore focus on the development of the nonintrusive authentication methods based on the users’ behavior in the past five years. Besides the authentication accuracy, we found that the usability and reliability of a smartphone user authentication system need to be considered. For example, in user register stage, the user will be boring and not want to use the system, if the user is requested to repeat the same actions as many times and collected time too long. Moreover, the authentication system must automatic update the predicted models, because the habitual behaviors of the smartphone user maybe change over time. The authentication accuracy may drop down if the predicted model does not keep the latest. In this three-year proposal, we plan to apply the active learning strategy to improve the model building in user register stage. In the second year, we plan to apply the online learning strategy to automatic update the predicted models. We plan to study the domain adaption of the transfer learning strategy and apply it to reduce the retraining time of the model updating in the third year.
