Information granulation via neural network-based learning

This paper concerns with an information granulation approach that is based on neural network learning. The approach involves three key phases. First, information granules are induced in the space of numerical data via a soft competitive learning algorithm with the ability to automatically determine the granularity level needed to properly model the data. Then, information granules are fuzzified, i.e. quantified in terms of fuzzy sets and used as building blocks of a fuzzy rule-based model. Finally, a supervised learning phase is applied to adjust the shape and the distribution of fuzzy granules. The approach is illustrated with the aid of a numerical example that provides insight into the validity of the induced granules and their effect on the results of computing.