Fuzzy cognitive map learning based on improved nonlinear Hebbian rule

Fuzzy cognitive map (FCM) is a powerful soft computing technique for modeling complex systems. It is a combination of fuzzy logic theory and neural networks. Developing of FCM is easy and adaptable based on human knowledge and experience. On the other hand, the main dependence on experts' knowledge and opinion, and the potential convergence to undesire steady states are the shortcomings of FCMs. Learning methods are good choices used to overcome the shortcomings and strengthen the efficiency and robustness of FCM. This paper proposes one improved Hebbian algorithm on non-linear units for training FCMs. With the proposed learning procedure, FCM can modify its fuzzy causal web as casual pattern change and update their causal knowledge as experts.