ESOINNとCounter Propagation Neural Networkを用いた追加学習可能なクラス分類法

カテゴリ: 論文誌(論文単位)

グループ名: 【C】電子・情報・システム部門

発行日: 2016/07/01

タイトル(英語): A Clastering Method for Incremental Learning using ESOINN and Counter Propagation Neural Networks

著者名: 河合　俊典（千葉工業大学大学院 情報科学研究科），山口　智（千葉工業大学 情報工学科）

著者名(英語): Shunsuke Kawai (Graduate School of Information and Computer Science, Chiba Institute of Technology), Satoshi Yamaguchi (Dept. of Computer Science, Chiba Institute of Technology)

キーワード: 自己組織化マップ，自己増殖型ニューラルネットワーク，ニューラルネットワーク，追加学習，カウンタプロパゲーション　　self-organizing Map，self-organizing Incremental Neural Networks，Neural Networks，Incremental Learning，Counter Propagation

要約(英語): This paper proposes a method for automatic labeling for incremental learning. In our method, an ESOINN (Enhanced Self-organizing Incremental Neural Network) and a counter propagation neural network are used. ESOINN is a neural network that copes with incremental learning. However, since the training of an ESOINN uses unsupervised learning, users have to label the input data based on the output of the ESOINN by hand. In our proposed method, output values of the ESOINN are used as input to the counter propagation neural network. The counter propagation neural network is trained by supervised learning. The desired output values of the counter propagation neural network are the label of the data that are corresponding to input to ESOINN. By using these neural networks, our method is able to label input data automatically. The proposed method was applied to two clustering problems: handwritten digit recognition and natural image recognition. In these applications, our method showed better performance for clustering and incremental learning than did an ESOINN alone.

本誌: 電気学会論文誌C（電子・情報・システム部門誌） Vol.136 No.7 （2016） 特集：平成27年電子・情報・システム部門大会

本誌掲載ページ: 945-954 p

原稿種別: 論文／日本語

電子版へのリンク: https://www.jstage.jst.go.jp/article/ieejeiss/136/7/136_945/_article/-char/ja/