Efficient Algorithms of Kernelized Hard c-Means Based on Cosine Correlation

• Inokuchi Ryo

  University of Tsukuba

• Mizutani Kiyotaka

  University of Tsukuba

• Miyamoto Sadaaki

  University of Tsukuba

Abstract

Kernelized clustering algorithms are successful to obtain nonlinear cluster boundaries. Among them, kernelized hard c-means based on cosine correlation is useful for the document classification. However, it has the drawback of a high computational effort when a kernel function is used. In this paper, we propose new time-efficient algorithms for kernelized hard c-means based on cosine correlation. Our approach is that on-line algorithms are kernelized instead of batch algorithms such as kernelized hard c-means. Numerical examples show the effectiveness of the proposed method.

Journal

• SCIS & ISIS

  SCIS & ISIS 2006 (0), 1743-1746, 2006

  Japan Society for Fuzzy Theory and Intelligent Informatics

Keywords

• Kernel Method
• Hard c-Means
• On-line Algorithms

Details

• CRID: 1390282680567176448
• NII Article ID: 130004672424
• DOI: 10.14864/softscis.2006.0.1743.0
• Text Lang: en
• Data Source:
  • JaLC
  • CiNii Articles
• Abstract License Flag: Disallowed