High speed associative memories for feature extraction and visualisation

  • Authors:
  • Aistis Raudys

  • Affiliations:
  • Data Analysis Department, Institute of Mathematics and Informatics, Akademijos 4, Vilnius LT-2600, Lithuania

  • Venue:
  • Pattern Recognition Letters
  • Year:
  • 2003

Quantified Score

Hi-index 0.10

Visualization

Abstract

New feature extraction method that handles nonlinearly separable datasets, preserves data geometry, minimises classification error directly and is designed especially for visualisation is suggested. The method employs hardware friendly binary correlation matrix memories (CMM), which makes the algorithm itself hardware friendly. To find coefficients of optimal linear orthogonal transformation and to speed up the calculations, binary CMM classifier and modified genetic optimisation technique are applied. The proposed technique was verified and compared with four competitive mapping techniques over a dozen of artificial and real world datasets. Experiments performed with respect to visualisation and classification accuracy showed that method is preferable to use on average sized nonlinear problems for extracting two features on behalf of visualisation.