최적 신경망 분류기 학습을 위한 PCA망의 성능고찰 = On Performance of PCA Networks for Optimal Neural-Network Classifiers Learning

다국어 초록 (Multilingual Abstract)

Neural network implementations of principal component analysis(PCA) provide a means for unsupervised feature discovery and dimension reduction. In this paper we report the empirical results o feature extractions via principal component analysis networks(PCA networks) in order to learn the optimal neural-network classifiers. Three PCA networks trained by generalized Hebbian algorithm(GHA), weighted subspace algorithm (WSA), and adaptive principal component extraction(APEX) are implemented to test their possibilities of using as a feature extractor for neural classifiers. Experimental results show that a significant reducton in network structure can be achieved without a sacrifice in classifier accuracy. Thus we verified that the PCA networks can be used as a viable tool for designing the optimal neural classifiers.