計測と制御

抄録を表示する抄録を非表示にする

Learning machines are noticed with interest as one of the important fields in the research of artificial intelligence in recent years.
This paper reports their basic description, results of some improvements on each part of a learning machine, its practical application and developed learning machines adopting new quantities in patterns, weight function and pattern components.
The learning machine is composed of three parts, that is, input part, adaptation part and decision part. In this paper, feature selection mechanism, non-linear weight elements and hysteresis decision elements are introduced in the input part, the adaptation part and the decision part, respectively, and their merits and theories are reported.
Results of vectorcardiographic diagnosis utilizing learning machine techniques are also reported as an example of the practical application of the learning machine, and it is concluded that it has fairly high information processing faculty inspite of its simple mechanism.
The learning machine is extended to have patterns and weight function given by continuous functions instead of the usual vectors with finite dimension. This learning machine is proved to be useful for solution of Fredholms integral equation of the first kind.
A new type learning machine is proposed whose input components and desired output are given by vector quantities and proof of convergence of its learning process is given. It is also showed that this new type machine has many application fields such as, automatic color matching computer, automatic colored area calculator and automatic gas-chromatogram analyser.

抄録全体を表示

PDF形式でダウンロード (2631K)

J-STAGEへの登録はこちら（無料）