Kernel approaches to unsupervised and supervised machine learning

Research output: Chapter in Book/Report/Conference proceedingConference contribution

14 Scopus citations


In the kernel approach, any N vectorial or non-vectorial data can be converted to N vectors with feature dimension N. The promise of the kernel approach hinges upon its representation vector space, leading to a "cornerized" data structure. Furthermore, the nonsingular kernel matrix basically assures a theoretically linear separability, critical to supervised learning. The main results are two folds: In terms of unsupervised clustering, the kernel approach allows dimension reduction in the spectral space and, moreover, a simple error analysis for the fast kernel K-means. As to supervised classification, by imposing uncorrelated perturbation to the training vector in the spectral space, a perturbed (Fisher) discriminant analysis (PDA) is proposed. This ultimately leads to a hybrid classier which includes PDA and SVM as specials cases, thus offering more flexibility for improving the prediction performance.

Original languageEnglish (US)
Title of host publicationAdvances in Multimedia Information Processing - PCM 2009 - 10th Pacific Rim Conference on Multimedia, Proceedings
Number of pages32
StatePublished - 2009
Event10th Pacific Rim Conference on Multimedia, PCM 2009 - Bangkok, Thailand
Duration: Dec 15 2009Dec 18 2009

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume5879 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Other10th Pacific Rim Conference on Multimedia, PCM 2009

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • General Computer Science


  • Dimension reduction
  • Error analysis
  • Fisher discriminant analysis(FDA)
  • Kernel matrix
  • PDA-SVM hybrid classifier
  • Perturbed discriminant analysis(PDA)
  • Scatter matrix
  • Spectral space
  • Support Vector Machine(SVM)


Dive into the research topics of 'Kernel approaches to unsupervised and supervised machine learning'. Together they form a unique fingerprint.

Cite this