Learning the Fuzzy Connectives of a Multilayer Network Using Particle Swarm Optimization

Fuzzy connectives provide a simple and yet a very flexible way to carry out multicriteria aggregation. Often in complex problems, the aggregation needs to be carried out at different hierarchical levels. Multilayer networks provide a natural and intuitive way for modeling such hierarchical decision making systems. In this paper we propose a novel, guided heuristic for learning the parameters of a multilayer network using particle swarm optimization. We also investigate the possibility of having multiple ways of aggregating the same information based on training data. Experiments are run by selecting several different topologies for the multilayer network. Also, a comparison is made between our method and another approach that uses back propagation for training