主動式學習應用於非侵入式智慧型手機驗證機制之使用者行為建模方法最佳化研究;Optimized Active Learning to Collect User’s Behavior for Training Model Based on Non-intrusive Smartphone Authentication

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/72137

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/72137

题名:	主動式學習應用於非侵入式智慧型手機驗證機制之使用者行為建模方法最佳化研究;Optimized Active Learning to Collect User’s Behavior for Training Model Based on Non-intrusive Smartphone Authentication
作者:	書瑪寧;Putri,Ika Kusumaning
贡献者:	資訊工程學系
关键词:	非侵入式識別;使用者識別;主動學習方法;支持向量機器;non-intrusive authentication;user authentication;active learning;support vector machine
日期:	2016-07-28
上传时间:	2016-10-13 14:27:48 (UTC+8)
出版者:	國立中央大學
摘要:	為了保護手機上的隱私資料，目前已存在數個識別機制，如：PIN碼、密碼解鎖，及生物特徵方式的識別機制。然而這些識別方法於便利性及安全性上人仍有所不足。非侵入式識別機制由於僅須收集使用者行為即能進行識別，故能彌補上述識別方式之不便利性。目前已存有數個非侵入式識別機制，但並沒有考慮訓練樣本的收集時間過長的問題。而以門檻為停止條件的主動學習方法雖能減少訓練樣本量，卻會造成錯誤率上升。　　於本研究中，我們提出一個優化後的主動學習方法，使其更為有效地收集訓練資料。支持向量機器被用來分析少量的訓練資料。本研究提出兩項主要的方法，其一為使用優化後的停止條件，藉以減少資料量。其二則為使用改善的模型分析方法決定訓練資料之來源，藉以保持其原有的錯誤率。　　於實驗後，我們發現本研究所提出方法相比原有主動學習方法有較好的效果。訓練資料收集時間可從17分鐘降至10分鐘，約減少至原所需時間量的41%，並保持相同的錯誤率。關鍵字:非侵入式識別，使用者識別，主動學習方法，支持向量機器 ;In order to protect the data in the smartphone, there is some protection mechanism that has been used. The current authentication uses PIN, password, and biometric-based method. These authentication methods are not sufficient due to convenience and security issue. Non-Intrusive authentication is more comfortable because it just collects user’s behavior to authenticate the user to the smartphone. Several non-intrusive authentication mechanisms were proposed but they do not care about the training sample that has a long data collection time. The Threshold-based active learning has proposed the method that cut down the training data but it makes the error rate increase. In this research, we propose a method to collect data more efficient using Optimized Active Learning. The Support Vector Machine (SVM) used to identify the effect of some small amount of training data. This proposed system has two main functionalities. First, to cut down the training data using optimized stop rule. Second, maintain the Error Rate using modified model analysis to determine the training data that fit for each user. Finally, after we done the experiment, we conclude that our proposed system is better than Threshold-based Active Learning. The time required to collect the data can cut down to 41% from 17 to 10 minutes with the same Error Rate. Keywords: non-intrusive authentication, user authentication, active learning, support vector machine
显示于类别:	[資訊工程研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	263	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....