Neural learning in automatic fuzzy systems synthesis

This paper presents a self-organizing neural structure with neuron relocation features. The neural net is used in the automatic synthesis of a dynamic self-organizing fuzzy system (DSOFS). The neural relocation learning provides a way to add, adapt and/or remove the fuzzy rules and the reference fuzzy sets of the DSOFS. The neural equivalent of modifying the DSOFS rules is adding and/or disposing the neurons while learning the input-output behaviour. This algorithm extends the topological ordering concept. A basin of attraction is supposed for every neuron (fuzzy rule) as a ground for the fuzzy reference sets construction. The DSOFS synthesis in a pattern recognition problem is showed.