Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Multi-Class Linear Dimension Reduction by Generalized Fisher Criteria

Marco Loog (2,1), Reinhold Haeb-Umbach (1)

(1) Philips Research Laboratories Aachen, Germany
(2) SSOR, Faculty of Information Technology and Systems, Delft University of Technology, The Netherlands

Linear Disciminant Analysis is in general unable to find the lower-dimensional feature space which maximizes the class discrimination, even if the class distributions can be assumed to be very simple, e.g. Gaussians with identical covariance matrices. In this paper we reformulate the K-class Fisher criterion as a sum of K(K-1)/2 two-class Fisher criteria. This formulation allows to weigh class pair contributions according to their relevance for classification. Further it offers an obvious way how to cope with heteroscedastic models. We propose a particular weighting scheme which attempts to approximate the pairwise Bayes error. Moderate improvements are obtained on the TIMIT phoneme classification task.


Full Paper

Bibliographic reference.  Loog, Marco / Haeb-Umbach, Reinhold (2000): "Multi-class linear dimension reduction by generalized Fisher criteria", In ICSLP-2000, vol.2, 1069-1072.