Learning of Neural Newtorks from Fuzzy Input and Fuzzy Output Data(Journal of Japan Society for Fuzzy Theory and Systems)

概要

論文の詳細を見る
This paper proposes a learning method of multilayer feedforward neural networks from fuzzy input vectors and fuzzy target outputs. The aim of this paper is to utilize linguistic information represented by fuzzy if-then rules as well as numerical data in the learning of neural networks. First we show an architecture of neural networks for dealing with input vectors of fuzzy numbers. A fuzzy input vector is mapped to a fuzzy number output by a neural network with the proposed architecture where the input-output relation of each unit is defined by the extension principle. Next we define a cost function by the fuzzy number output from the neural network and the corresponding target output that is also a fuzzy number. The cost function can be viewed as the difference between the two fuzzy numbers. We derive a learning algorithm from the cost function in the same manner as the BP algorithm. We also propose learning methods that are based on the inclusion relations between the fuzzy actual output and the fuzzy target output. The proposed learning methods are illustrated by simulation results for numerical examples.
日本知能情報ファジィ学会の論文
1992-10-15