木構造を有する階層型確率ニューラルネットの提案と指形状識別への応用

抄録

This paper proposes a new pattern classification method using probabilistic neural networks with hie rarchical structure. In this method, a log-linearized Gaussian mixture networkas a probabilistic neural networki sused at each node of a herarchical classification tree. The method proposed here automatically constructs a tree of neural networks from given data, and can achieve suitable discrimination. Also, Cross-validation for a constructed neural tree can improve the generalization ability for discrimination of unlearned dat a. Validity of the proposed method is demonstrated with discrimnation of hand shape.