Authors: Michal Gregor,Peter Groumpos
The paper proposes a new learning method for fuzzy cognitive maps, which makes it possible to encode an attractor into the map. The method is based on the principle of backpropagation through time known from the theory of artificial neural networks. Simulation results are presented to show how well the method performs. It is shown that the results are superior to those achieved using Hebbian learning approaches such as nonlinear Hebbian learning. Some lines for possible future research and development are given.