IDENTIFICATION ALGORITHMS FOR CHOQUET INTEGRAL AGENT NETWORKS
Satoru Iwasaki, Nakamura Kazuo · 2007
Choquet Integral Agent Networks (CHIAN) are proposed as methods realizing flexible information fusion mechanisms. In case of multi-layered network structures the conceptual representation of the information fusion as an input-output system is mathematically expressed as: y = F(u; w) where u is a vector at the input layer, y is a vector at the output layer and w is a vector composing fuzzy measures between adjacent layer agents from lower level to upper level. This paper is addresses to two types of identification problems, i.e. system identification and input identification, under specific constraints. Then the conceptual learning algorithms proposed in the authors' previous studies similar to the back propagation methods employed in the Neural Network theories were implemented into practical computer programs, And the validity of these specific algorithms are illustrated through some numerical experiments in the several types of problems with essential CHIAN structures, logical functions, and continuous linear/nonlinear functions.