Machine learning of neuro-fuzzy system based on fuzzy truth value

The paper addresses the problem of computational complexity of fuzzy inference. An outline of common fuzzy inference methods is provided. Since these methods cannot be used in case of multiple fuzzy inputs due to exponential computational complexity, another inference method is introduced. This method is based on fuzzy truth value, and it is free of the described drawback. The operation of the method is described further. The article provides a formal definition of the method for a single fuzzy rule as well as for multiple rules. Then, the issue of machine learning of such systems is concerned, including the parameters that can undergo adjustment and objective function. In the article, training by means of evolution strategy $(\mu, \lambda)$ is assumed. Finally, the described approach is being assessed in terms of approximation quality of a specific known function.