Structures of the covariance matrices in the classifier design

  • Šarūnas Raudys
  • Aušra Saudargiene
Statistical Classification Techniques
Part of the Lecture Notes in Computer Science book series (LNCS, volume 1451)


Structurization of the covariance matrices helps to reduce a number of parameters to be estimated. Own assumptions on the structure of the matrix are correct the structurization of the covariance matrix helps to reduce the generalization error in small learning-set cases. Efficacy of the matrix structurization increases if one decorrelates and scales the data, and uses the optimally stopped single layer perception classifier afterwards.

Index terms

regularized discriminant analysis learning-set size generalization dimensionality covariance matrix parameters reduction single layer perception 


  1. Deev, A.D. (1974). Discriminant function designed on independent blocks of variables.-Proc. Acad. of Sci. of USSR, Eng. Cybernetics, (USSR J.), 12,. 153–156 (in Russian).Google Scholar
  2. Friedman J.M. (1989). Regularized discriminant analysis. J. American Statistical Association, 84, 165–175.Google Scholar
  3. McLachlan, G.J.(1992). Discriminant Analysis and Statistical Pattern Recognition. Willey.Google Scholar
  4. Mcshalkin, L.D., & Serdobolskij, V.I. (1978). Errors in classifying multivariate observations. Theory of Probabilities and Applications, 23(4), 772–781 (in Russian)Google Scholar
  5. Raudys, S. (1972). On the amount of a priori information in designing the classification algorithm. Proc. Acad. of Sci. of USSR, Eng. Cybernetics, 14, 168–174 (in Russian).Google Scholar
  6. Raudys, S. (1991). Methods for overcoming dimensionality problems in statistical pattern recognition. A review. Zavodskaya Laboratorya (Factory Lab., USSR Journal), Moscow: Nauka, 3, 45 & 49-55 (in Russian).Google Scholar
  7. Raudys, S. (1996). Linear classifiers in perception design, ICPR13, Proc. 13th Int. Conf. on Pattern Recognition (Vienna, Austria, Aug.25–29) Vol. 4, Track D: Parallel and Connectionist Systems, IEEE Computer Society Press, Los Alamitos, 1996, 763-767.Google Scholar
  8. Raudys, S, (1998), Evolution and generalization of a single neurone. Part l. SLP as Seven statistical classifiers. Neural Networks (accepted).Google Scholar
  9. Raudys, S. & Jain, A.K. (1991). Small sample size effects in statistical pattern recognition: Recommendations for practitioners-IEEE Trans. on Pattern Analysis and Machine Intelligence, PAMI-13, 252–264.CrossRefGoogle Scholar
  10. Raudys, S. & Pikelis, V. (1980). Ondimensionality, sample size, classification error and complexity of the classification algorithm in pattern recognition. IEEE Trans. on Pattern Analysis and Machine Intelligence, PAMI-2 (3), 242–252.Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1998

Authors and Affiliations

  • Šarūnas Raudys
    • 1
  • Aušra Saudargiene
    • 1
  1. 1.Institute of Mathematics and InformaticsVilnius

Personalised recommendations