INCREMENTAL AND REGULARIZED LINEAR DISCRIMINANT ANALYSIS

Abstract

Linear discriminant analysis (LDA) is a well-known supervised dimensionality reduction technique, which has been applied successfully in many important applications such as pattern recognition, information retrieval, face recognition, micro-array data analysis and text classification. In this thesis, by deriving the mathematical relationship between orthogonal linear discriminant analysis (OLDA) and regularized orthogonal linear discriminant analysis (ROLDA), we find a mathematical criterion for selecting the regularization parameter in ROLDA. Unlike other regularized LDA methods, no candidate set of regularization parameter is needed in our new proposed method. In addition, an LDA-based incremental dimensionality reduction algorithm, ILDA/QR, has been developed, in which bursts of data that contains new classes is being inserted in the form of random chunks (one by one or chunk by chunk).