首页    期刊浏览 2025年05月26日 星期一
登录注册

文章基本信息

  • 标题:ニューラルネット形成時の学習領域の記憶と知識獲得
  • 本地全文:下载
  • 作者:浦 環 ; 石井 和男
  • 期刊名称:日本造船学会論文集
  • 印刷版ISSN:0514-8499
  • 电子版ISSN:1884-2070
  • 出版年度:1992
  • 卷号:1992
  • 期号:171
  • 页码:587-593
  • DOI:10.2534/jjasnaoe1968.1992.587
  • 出版社:The Japan Society of Naval Architects and Ocean Engineers
  • 摘要:

    Learning samples, which is applied to generation of a neural network through the supervised learning, cannot cover all possible input vectors, especially when the number of neurons in the input layer is not small. The neural network, therefore, should be used accompanied with the description denoting the valid region of input vectors based on the distribution of learning samples. In this paper, the Descriptive Neural Network is introduced to make clear whether an input vector to a neural network is included in the learned samples. The output of the Descriptive Neural Network represents liability of the input vector to have been learned. Using the Descriptive Neural Network, knowledge implemented in a neural network can be enhanced by additional learning without losing previous knowledge. The advantage of the proposed system is demonstrated with a feed forward model network which represents the dynamics of a cruising type autonomous underwater vehicle.

国家哲学社会科学文献中心版权所有