A gradient descent learning algorithm for fuzzy neural networks

In order to train fuzzy neural nets fuzzy number weights have to be adjusted. Since fuzzy arithmetic automatically leads to monotonic increasing outputs a direct fuzzification of the backpropagation method does not work. Therefore, other strategies like evolutionary algorithms are being considered in the literature. In this paper we suggest a backpropagation based method of adjusting the weights. Furthermore, we show that by using the proposed method convergence can be guaranteed