F-007 An Efficient Construction of RBF Network Based on Training by SOM

• Item type: FIT(1)
• 公開日: 2008-08-20
• タイトル / 言語: en
• タイトル: F-007 An Efficient Construction of RBF Network Based on Training by SOM
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_5794
• 資源タイプ: conference paper
• 著者所属: 岩手県大
• 著者所属: 岩手県大
• 著者所属: 岩手県大
• 著者所属(英): en
• 著者所属(英): Dept. of Software and Information Science, Iwate Prefectural University
• 著者所属(英): en
• 著者所属(英): Dept. of Software and Information Science, Iwate Prefectural University
• 著者所属(英): en
• 著者所属(英): Dept. of Software and Information Science, Iwate Prefectural University
• 著者名: 山下, 和彦; / 水野, 佑治
• 著者名(英): Yamashita, Kazuhiko; Chakraborty, Goutam; Mizuno, Yuji
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: In this work we proposed to train both input to hidden as well as hidden to output layer weights of a RBF network using Self-Organizing Maps (SOM) training. Initially a two dimensional SOM is trained. Once SOM training is over, the SOM input to output weights determine the RBF hidden units' location in problem feature space. Next, for every individual sample the winner SOM output is identified, and the label (class) of the sample is tagged with SOM outputs. This tag information is used to decide the connection weights between RBF middle layer nodes to output nodes. Thus by just executing SOM the RBF network is constructed. The performance is compared with multilayer Perceptron trained with error back propagation.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA1242354X
• 書誌情報: 情報科学技術フォーラム講演論文集; 巻 7, 号 2, p. 325-326, 発行日 2008-08-20
• 出版者 / 言語: ja
• 出版者: 情報処理学会