Structural learning of neural networks for analysis and knowledge extraction

Iizaka Tatsuya, Matsui Tetsuro

doi:10.1541/ieejeiss1987.121.3_673

Bibliographic Information

Other Title

解析と知識の抽出のためのニューラルネットワークの構造化学習
カイセキトチシキノチュウシュツノタメノニューラルネットワークノコウゾウカガクシュウ

Search this article

Description

This paper presents structure of a neural network for analysis and knowledge extraction from continuous valued problems and a training method for the structure. The proposed neural network consists of two types of hidden units. One type of hidden units has weights connected to only a group of input units. Another one has weights connected to all input units. The former type of hidden units allows to analyze each relation between a certain input data and corresponding output data. The latter type of hidden units ensures performance of the neural network as same as conventional neural networks. While the neural network is been training, needless hidden units are pruned automatically by the superposed energy function, the structure learning with forgetting, and the compact structuring algorithm. Therefore, it is easy to analyze the neural network and extract knowledge.<br> The effectiveness of the proposed method is shown by function approximation, peak load forecasting of electric power and water flow forecasting into the dam.

Journal

IEEJ Transactions on Electronics, Information and Systems

IEEJ Transactions on Electronics, Information and Systems 121 (3), 673-680, 2001

The Institute of Electrical Engineers of Japan

Keywords

Details 詳細情報について

Export