Fuzzy gated neuronal architecture for pattern recognition

In this paper, a novel fuzzy gated neuronal architecture capable of utilizing all possible combinations of the decision planes between the points represented by weights in n-dimensional feature space is proposed. This is achieved by selecting a set of nodes based on an eligibility criteria and then letting these selected nodes to compete. The time sequence of winning nodes generated by a time-varying eligibility criterion provides a time-indexed expert opinions in respect of the class membership grades. These opinions when combined properly enhance the class label prediction accuracies to a great extent. The architecture is built on a fuzzy gated neuron model and a set of gate control functions. In addition, it is shown that the training of weights to represent the cluster centroids is not necessary resulting in a quick network set up time. The classification performance of the proposed network on a difficult 12-class synthetic 3-D object recognition problem indicates excellent results