Hierarchical backward chaining neuro-fuzzy model for structural adjustment of the classification rules

Abstract

An adaptive approach to structural tuning of fuzzy classification knowledge bases built on trend relations or rules and inverse logic inference is developed. Causes – effects interconnection is modelled using fuzzy relational equations with the hierarchical max-min/min-max composition. The hierarchical neuro-fuzzy model of inverse inference based on trend rules is proposed. The network allows simplifying the training process in comparison with the extended neuro-fuzzy network based on trend relations. Resolution of the problem of inverse inference is done using recurrent correlations, which correspond to adjustment of the coordinates of maximum of input terms membership functions and causes combinations significance measures for the expert solutions of the trend system of equations.