RECURSIVE LEAST-SQUARES APPROACH TO COMBINING PRINCIPAL AND MINOR COMPONENT ANALYSES

A novel approach for high-performance data compression using neural ne tworks is proposed. After the principal components of the input vector s are extracted, the error covariance matrix obtained in the recursive least square training process is used to perform minor components pru ning so that a higher compression ratio is achieved. Simulation result s show that our method effectively combines principal and minor compon ent analyses.