A development of granular logic neural networks

This paper proposes a way to develop a special type of fuzzy logic network-a granular logic network which generalizes the conventional fuzzy logic network by means of expanding the weights (and biases). Information granularity provides some flexibility on the determination of weights of a network. Five protocols are discussed here to realize the granular weights. The optimization of levels of granularities for different weights is done by an effective tool-particle swarm optimization. An illustrative simple example is given to show the study process of our approach.