빠른 학습을 위한 1-Slack One-Class SVM = 1-Slack One-Class SVM for Fast Learning

국문 초록 (Abstract)

One-class SVM은 데이터가 존재하는 영역을 추출하고, 이 영역을 서포트 벡터로 표현하며 표현된 영역 밖의 데이터들은 아웃라이어(outlier)로 간주된다. 본 논문에서는 One-class SVM의 성능은 그대로 유지하면서 대용량의 학습데이터에서 빠른 학습을 하기 위한 1-slack one-class SVM을 제안한다. 실험결과 One-class SVM과 1-slack one-class SVM 수행 결과가 거의 같으며, 대용량의 학습데이터에서 1-slack one-class SVM이 일반 One-class SVM보다 학습속도가 훨씬 빠름을 알 수 있었다.

다국어 초록 (Multilingual Abstract)

Training one-class support vector machine (one-class SVM) involves solving a quadratic programming (QP) problem. By increasing the number of training samples, solving this QP problem becomes intractable. In this paper, we reformulate a standard one-class SVM and derive a 1-slack one-class SVM. The 1-slack oneclass SVM fixes the bias term ρ =1 and replaces the n cutting-plane models of the hinge-loss with a single cutting plane model for the sum of the hinge-losses. Experimental results showed that the 1-slack one-class SVM is much faster than the standard one-class SVM without loss of performance.

목차 (Table of Contents)

• 요약
• Abstract
• 1. 서론
• 2. One-class SVM
• 3. 1-slack one-class SVM
• 4. 실험
• 5. 결론
• 참고문헌

참고문헌 (Reference)

1 D.M.J. Tax, "Support vector domain description" 54 : 45-66, 2004

2 D. Tax, "Online SVM Learning: from Classification and Data Description and Back" (1) : 499-508, 2003

3 Y. Zhang, "One-class classification of text streams with concept drift" 2008

4 L.M. Manevitz, "One-class SVMs for document classification" 2 : 139-154, 2001

5 Y. Chen, "One-class SVM for learning in image retrieval" 2001

6 H. Deng, "Model selection for anomaly detection in wireless ad hoc networks" 540-546, 2007

7 C.C. Chang, "LIBSVM: a library for support vector machines" 2 (2): 27:1-27:27, 2011

8 J. Ma, "Identifying suspicious URLs: An application of large-scale online learning" 681-688, 2009

9 B. Scholkopf, "Estimating the support of a high-dimensional distribution" 13 (13): 1443-1471, 2001

10 T. Joachims, "Cutting-Plane Training of Structural SVMs" 77 (77): 2009

11 A. Tavakkoli, "A support vector data description approach for background modeling in videos with quasi-stationary backgrounds" 17 (17): 635-658, 2008

12 J.J. Hull, "A database for handwritten text recognition research" 16 (16): 550-554, 1994