Expert- Independent Initial Weights Assignment Technique for Learning Fuzzy Cognitive Map With Hebbian Learning Algorithms

Abstract

Abstract Fuzzy Cognitive Maps (FCMs) are single layer neural network-like Supervised Machine Learning Algorithm which can be used as a tool for modelling dynamic systems in a graphical cause-effect relationships form. Though a powerful tool, FCM do not always converge to a desired state but relies on other learning algorithms to find connection matrix that will lead the system to a stable state. Of the various algorithms for learning FCM connection matrix, Hebbian variants are the earliest and simplest. However, they depend on human experts for initial weight matrix before they can be applied or learning commences. So without human experts, these algorithms are limited and even with availability of human expert, the weights could be a result of experts’ subjective opinion or limited knowledge of the system. This paper present scientific method and algorithm for finding initial weight from node activation values and guide in the choice of map density. This allow application of Hebbian learning algorithms without human experts and a multi-map and multi-density Hebbian Learning solution of FCM which may provide optimal solution than single map expert initial map