A new approach to multiple class pattern classification with random matrices.pdfVIP

A NEW APPROACH TO MULTIPLE CLASS PATTERN CLASSIFICATION WITH RANDOM MATRICES DONG Q. WANG AND MENGJIE ZHANG Received 8 April 2003 and in revised form 2 December 2003 We describe a new approach to multiple class pattern classification problems with noise and high dimensional feature space. The approach uses a random matrix X which has a specified distribution with mean M and covariance matrix r (Σ + Σ ) between any two ij s columns of X . When Σ is known, the maximum likelihood estimators of the expectation M , correlation Γ, and covariance Σs can be obtained. The patterns with high dimensional features and noise are then classified by a modified discriminant function according to the maximum likelihood estimation results. This new method is compared with a multi- layer feed forward neural network approach on nine digit recognition tasks of increasing difficulty. Both methods achieved good results for those classification tasks, but the new approach was more effective and more efficient than the neural network method for dif- ficult problems. 1. Introduction This paper presents a new approach to the multiple class pattern classification process. Consider the situation where the multivariate observation xi with p dimensions on an object consists of two independent components y i and εi (i = 1,2,... , n), where n is total sample size. Let y i have a p -dimensional multivariate normal distribution with mean vector µ and covariance matrix Σs , while εi has a multivariate normal distribution with mean vector 0 and covariance Σε . If we let x = y + ε and the vector x has a multivariate normal distribution with mean i i i