ASAFES.2: a novel, neuro-fuzzy architecture for fuzzy computing based on functional reasoning

Abstract

The proposed architecture, ASAFES2, is a function approximator which combines the functional reasoning or Sugeno's fuzzy reasoning method with stochastic reinforcement learning-a class of quite powerful neural network training algorithms. It is a simple and versatile mathematical tool for fuzzy computing, featuring smooth and quick convergence and ease of use. The main ideas are the fuzzy partitioning of the input space into fuzzy subspaces (each corresponding to a possible fuzzy rule), and the use of a separate, stochastic reinforcement learning neural unit (ANASA II) for every fuzzy subspace, in order to calculate the optimum consequence parameters. Some preliminary results are presented, proving ASAFES2 superior over backpropagation. A new, and "flexible" membership function is also proposed.<>