Summary
International Symposium on Nonlinear Theory and its Applications
2009
Session Number:A2L-D
Session:
Number:A2L-D1
Robust Laplacian Eigenmaps for Semi-Supervised Pattern Classification
Weiwei Du, Kiichi Urahama,
pp.-
Publication Date:2009/10/18
Online ISSN:2188-5079
DOI:10.34385/proc.43.A2L-D1
PDF download (157.2KB)
Summary:
The Laplacian eigenmap (LEM) is a popularly used graph spectral algorithm for mapping data nonlinearly into a low-dimensional space. In this paper, we robustify LEM and use it for semi-supervised pattern classification. The distance metric for feature vectors is modulated with a semi-supervised feature selection method and data are mapped into a low-dimensional classification space with the robustified LEM. Test data are classified by the nearest neighbor rule with the modulated distance metric between data.