PARAMETER ESTIMATION OF RBF NETWORKS FOR PATTERN CLASSIFICATION AND PDF ESTIMATION

概要

論文の詳細を見る
Pattern classification is important technique in a wide variety of field. Approaches to pattern classification are classified into boundary forming and probability density function (PDF) estimation. In another aspect, there exist supervised and unsupervised learning algorithms. The PDF estimation can be applied to pattern classification, however, classification error is not used as a cost function. The boundary forming approaches trained by the supervised learning using the targets cannot estimate the PDF. Because, the targets are independent of the PDF. In this paper, the relation between the pattern classification and the PDF estimation is investigated based on the RBF networks. Parameter estimation of the RBF network basically follows the PDF estimation.
社団法人電子情報通信学会の論文
1997-03-06