A neural net topology for bidirectional fuzzy-neuro transformation

In this paper, we propose an integrated neuro-fuzzy system (INFS) that facilitates the functional equivalent conversion between fuzzy systems and neural networks thus combining the advantages of both paradigms. The basis for the INFS constitutes a special neural network architecture with a structure corresponding to that of a fuzzy model. In a repeated cycle, knowledge acquired from an expert is converted from a fuzzy system to a neural net which is applied to a target system to learn from the data. After completed adaptation the neural network is translated back into a fuzzy model. First results demonstrate the significant performance with respect to data-driven optimization of fuzzy system components.<>