Learning fuzzy cognitive maps using decomposed parallel ant colony algorithm and gradient descent

Fuzzy cognitive maps (FCMs) are a model for causal modeling and causal inference. It represents the real-world concepts and the causal relations between the concepts by using fuzzy variables. The major benefit of the fuzzy variables is that the model is more robust to the errors in the observed data. Although FCMs have been widely used in different research areas, it is still an open problem to efficiently construct large scale FCM models. To further improve the efficiency of the existing FCM learning algorithms, we propose a new algorithm that combines ant colony optimization algorithm, gradient descent local search and a decomposed parallel computing framework to build large scale FCMs from observational data. A set of network inference problem is used to evaluate the performance of the proposed algorithm and the results are compared to other algorithms including traditional ant colony optimization, and real coded genetic algorithms. Experimental results suggest that our algorithm outperforms the other algorithms in terms of model accuracy. We also compared the computation time required by the non-parallel ant colony optimization algorithm and the proposed parallel algorithm. When the number of nodes is appropriate, the speedup could be very close to linear speedup.