Premium
Complexity versus integrity solution in adaptive fuzzy‐neural inference models
Author(s) -
Dimirovski Georgi M.
Publication year - 2008
Publication title -
international journal of intelligent systems
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.291
H-Index - 87
eISSN - 1098-111X
pISSN - 0884-8173
DOI - 10.1002/int.20283
Subject(s) - adaptive neuro fuzzy inference system , computer science , inference , artificial intelligence , artificial neural network , neuro fuzzy , fuzzy control system , a priori and a posteriori , fuzzy logic , minification , membership function , machine learning , function (biology) , data mining , philosophy , epistemology , evolutionary biology , biology , programming language
This paper explores aspects of computational complexity versus rule reduction and of integrity preservation versus optimality index, which have become an issue of considerable concern in learning techniques for adaptive fuzzy inference models. In control‐oriented applications of adaptive fuzzy inference systems, implemented as fuzzy‐neural networks, a balanced observation of these conflicting requirements appeared rather important for a good yet feasible application design. The focus is confined to a family of adaptive fuzzy inference systems that can be interpreted as a partially connected multilayer feedforward neural networks employing Gaussian activation function. The knowledge base rules are designed implying the connections are a priori fixed, and then the respective strengths adapted on the grounds of input and output data sets. Information granulation plays a significant role too. These as well as membership‐function parameters ought to be adapted in a learning‐training process via the minimization of an appropriate error function. © 2008 Wiley Periodicals, Inc.