Abstract for HONS 04/14 - Computer Science and Software Engineering - University of Canterbury - New Zealand

Abstract for HONS 04/14

Multi-Centroid PSO Classification Learning on the GPU

Cain Cresswell-Miley
Department of Computer Science and Software Engineering
University of Canterbury

Abstract

Training classifiers can be seen as an optimization problem. With this view, we have developed a method to train a type of nearest centroid classifier with PSO. Results showed an improvement on most of the datasets tested. Additionally, we have developed a method to utilize the developed classifier with datasets containing both numeric and categorical data by integrating the centroid algorithm with a decision tree. However, experiments found no significant improvement over the original decision tree method. Both the developed PSO centroid algorithm, and the previous PSO centroid algorithm are implemented on the GPU, with results showing at least one order of magnitude difference between speeds of the GPU and a ‘typical’ sequential CPU implementation.
  • Phone: +64 3 369 2777
    Fax: +64 3 364 2569
    CSSEadministration@canterbury.ac.nz
  • Computer Science and Software Engineering
    University of Canterbury
    Private Bag 4800, Christchurch
    New Zealand
  • Follow us
    FacebookYoutubetwitterLinked In