Linguistic neurocomputing: the design of neural networks in the framework of fuzzy sets

A process of information granulation takes care of an enormous flood of numerical details that becomes summarized and hidden (encapsulated in the form of fuzzy sets) at the time of the design of a neural network. Information granules play an important role in the development of neural networks. First, they substantially reduce the amount of training as the designed network needs to deal with a significantly reduced and highly compressed number of data that falls far below the size of the original training set. The same granulation mechanism delivers some highly advantageous regularization properties. Second, information granules support the design of more transparent and easily interpretable neural networks. The necessary effect of information granulation is accomplished in the framework of fuzzy sets, especially via context-sensitive (conditional) fuzzy clustering. Subsequently, the resulting neural network becomes an architecture with nonnumeric connections. A thorough analysis of results of computing carried out in the setting of linguistic neurocomputing is also given. [Copyright &y& Elsevier]