Skip navigation

DSpace

機構典藏 DSpace 系統致力於保存各式數位資料(如:文字、圖片、PDF)並使其易於取用。

點此認識 DSpace
DSpace logo
English
中文
  • 瀏覽論文
    • 校院系所
    • 出版年
    • 作者
    • 標題
    • 關鍵字
    • 指導教授
  • 搜尋 TDR
  • 授權 Q&A
    • 我的頁面
    • 接受 E-mail 通知
    • 編輯個人資料
  1. NTU Theses and Dissertations Repository
  2. 電機資訊學院
  3. 電機工程學系
請用此 Handle URI 來引用此文件: http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/69318
完整後設資料紀錄
DC 欄位值語言
dc.contributor.advisor于天立
dc.contributor.authorHao-Lun Hsuen
dc.contributor.author許浩倫zh_TW
dc.date.accessioned2021-06-17T03:12:48Z-
dc.date.available2020-07-16
dc.date.copyright2018-07-16
dc.date.issued2018
dc.date.submitted2018-07-13
dc.identifier.citation[1] Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. ”Why should I trust you?”: Explaining the predictions of any classifier. In Proceedings of the 22Nd ACM SIGKDD International Conference on Knowledge Discovery and Data Min- ing, KDD ’16, pages 1135–1144, New York, NY, USA, 2016.
[2] Steven L. Salzberg. C4.5: Programs for machine learning by j. ross quinlan. morgan kaufmann publishers, inc., 1993. Machine Learning, 16(3):235–240, Sep 1994.
[3] D.T. Pham and M.S. Aksoy. Rules: A simple rule extraction system. Expert Systems with Applications, 8(1):59 – 65, 1995.
[4] Shai Shalev-Shwartz and Shai Ben-David. Understanding machine learning: From theory to algorithms. Cambridge university press, 2014.
[5] Vladimir Vapnik, Esther Levin, and Yann Le Cun. Measuring the VC-dimension of a learning machine. Neural Comput., 6(5):851–876, 1994.
[6] V. N. Vapnik and A. Ya. Chervonenkis. On the uniform convergence of relative frequencies of events to their probabilities. Theory of Probability & Its Applications, 16(2):264–280, 1971.
[7] Tom M. Mitchell. Version spaces: A candidate elimination approach to rule learning. In Proceedings of the 5th International Joint Conference on Artificial Intelligence - Volume 1, IJCAI’77, pages 305–310, San Francisco, CA, USA, 1977. Morgan Kauf- mann Publishers Inc.
[8] Tom M. Mitchell. Generalization as search. Artificial Intelligence, 18(2):203 – 226,1982.
[9] Andrzej Ehrenfeucht, David Haussler, Michael Kearns, and Leslie Valiant. A gen- eral lower bound on the number of examples needed for learning. Information and Computation, 82(3):247 – 261, 1989.
[10] Yoav Freund, H. Sebastian Seung, Eli Shamir, and Naftali Tishby. Selective sampling using the query by committee algorithm. Machine Learning, 28(2):133–168, Aug1997.
[11] Maria-Florina Balcan, Steve Hanneke, and Jennifer Wortman Vaughan. The true sample complexity of active learning. Machine Learning, 80(2):111–139, Sep 2010.
dc.identifier.urihttp://tdr.lib.ntu.edu.tw/jspui/handle/123456789/69318-
dc.description.abstract隨著機器學習的技術快速發展,機器學習在解決現實世界的問題扮演重要的角色。然而,為了使模型的能力更強大,學習模型因此變得越來越複雜,人類也因此越來越難理解模型會如何決策,或者在特定情況下會做出什麼判斷。由於對學習模型的不理解,人類會因而對模型產生不信任。因此,在這篇論文中,我定義了可驗證度和可驗證維。前者是一個衡量至少多少比例的輸入空間的點可以被所有模型空間中的模型正確預測。越高的可驗證度表示有越高的機率,模型會在其他的點上做出正確的預測,我們也能因此越信任模型。後者則是在衡量對於一個模型空間,其需要多少的訓練維度,才能確保平均的驗證度大於零。除此之外,在本論文中,我推導出驗證維的下界是等效VC維 (VC維的一種變體,其考慮了輸入空間的分佈),還有可驗證度的上下界(上界與等效VC維有關、下界則與驗證維有關)。最後,我討論了可驗證度的取樣複雜度,並且證明了不可驗證度的下界是模型空間中最大的真實誤差。zh_TW
dc.description.abstractThe development of machine learning has grown rapidly and has been successfully applied to many real-world applications. However, the learning models are less comprehensible by humans. Such incomprehension causes distrust. I believe that for common users, comprehension comes from how a model behaves under certain circumstances. Thus, in this thesis, I propose the verifiability and the verifiability dimension. The former is an indicator of what proportion of instances are correctly classified by all hypotheses. Those hypotheses are generated by different learning algorithms given the same hypothesis space and the same training instances. With the higher verifiability, users can know that models make a correct prediction on an unknown instance with higher probability. The latter is the complexity of a hypothesis space that the average verifiability is greater than zero given the training instances. Besides, I show that the lower bound on the verifiability dimension is the effective Vapnik-Chervonenkis (VC) dimension, another version of VC dimension which takes the distribution of instance space into account. Also, I show that the upper and lower bounds on the average verifiability in terms of the effective VC dimension and the verifiability dimension. Finally, I discuss the sample complexity of unverifiability, the minimal number of training instances needed to ensure the unverifiability small enough. The lower bound on unverifiability is the maximal true error of all hypotheses in the version space. As a result, in statistical learning, the sample complexity of unverifiability is no less than the sample complexity of true error, which is $Theta(frac{VC_H}{epsilon})$.en
dc.description.provenanceMade available in DSpace on 2021-06-17T03:12:48Z (GMT). No. of bitstreams: 1
ntu-107-R05921048-1.pdf: 3173233 bytes, checksum: 0df7cf93b436ba1af8299374942fd942 (MD5)
Previous issue date: 2018
en
dc.description.tableofcontents口 試 委 員 會 審 定 書 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
致 謝 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
摘 要 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Preliminary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 Learning settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Version Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Verifiability and Verifiability Dimension . . . . . . . . . . . . . . . . . . . 7
3.1 The verifiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 The verifiability dimension . . . . . . . . . . . . . . . . . . . . . . . . .10
3.3 The Vapnik–Chervonenkis (VC) dimension and the effective VC dimension 14
3.3.1 Dichotomy and shatter . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3.2 The VC dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3.3 The Effective VC dimension . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Lower bound on the verifiability dimension . . . . . . . . . . . . . . . 21
3.5 Lower and upper bounds on average verifiability . . . . . . . . . . . . 23
4 Sample Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.1 Sample complexity of random sampling algorithm . . . . . . . . . . . . 33
4.2 Relation between unverifiability and true error . . . . . . . . . . . . . . 35
4.3 Lower bound on sample complexity of the unverifiability . . . . . . . . .37
5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
dc.language.isoen
dc.subject等效VC維zh_TW
dc.subject可驗證度zh_TW
dc.subject可驗證維zh_TW
dc.subjectVC維zh_TW
dc.subjectVC dimensionen
dc.subjecteffective VC dimensionen
dc.subjectverifiability dimensionen
dc.subjectverifiabilityen
dc.title以驗證點和可驗證度的觀點討論機器學習理論zh_TW
dc.titleVerified Instances and Verifiability: An Alternative Theoretical Prospective of Learningen
dc.typeThesis
dc.date.schoolyear106-2
dc.description.degree碩士
dc.contributor.oralexamcommittee陳和麟,王奕翔
dc.subject.keyword可驗證度,可驗證維,VC維,等效VC維,zh_TW
dc.subject.keywordverifiability,verifiability dimension,VC dimension,effective VC dimension,en
dc.relation.page42
dc.identifier.doi10.6342/NTU201801408
dc.rights.note有償授權
dc.date.accepted2018-07-13
dc.contributor.author-college電機資訊學院zh_TW
dc.contributor.author-dept電機工程學研究所zh_TW
顯示於系所單位:電機工程學系

文件中的檔案:
檔案 大小格式 
ntu-107-1.pdf
  未授權公開取用
3.1 MBAdobe PDF
顯示文件簡單紀錄


系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。

社群連結
聯絡資訊
10617臺北市大安區羅斯福路四段1號
No.1 Sec.4, Roosevelt Rd., Taipei, Taiwan, R.O.C. 106
Tel: (02)33662353
Email: ntuetds@ntu.edu.tw
意見箱
相關連結
館藏目錄
國內圖書館整合查詢 MetaCat
臺大學術典藏 NTU Scholars
臺大圖書館數位典藏館
本站聲明
© NTU Library All Rights Reserved