Research on Higher Order Artificial Neuron for E-Information Systems

In this paper, a novel model based on Complex Valued Single Neurons which make the composition of MLP and RBf, aggregate their response linearly and non-linearly in two new proposed single neurons respectively is presented. The learning and generalization capabilities of these neurons have been tested over various benchmark and real life problems. Since complex numbers have natural representation of phase and magnitude that locates the complex number uniquely on the plane, these neurons have shown their potent computational power over wide spectrum of geometrical transformations. The artificial neural network based applications in various enterprise information systems are growing interest for comprehensive coverage and understanding of organizational competitiveness. The neurobiologists studies have observed various ways for interaction of synaptic inputs in the dendritic tree which eventually reflect the computation performed by a neuron. The analysis of aggregation linearly or non-linearly has been a subject of interest for NN researchers. Most of the aggregated neuron models are based on ‘Multiplication’ in an attempt to consider non-linearity in ANN. Prior research work pertaining to implementation of the input derived from the relationship among the second and higher order polynomial as a set of inputs is available. The novel model presented in this paper has the potential to improve the ANN performance by virtue of faster response achieved as a result of aggregated input constituted of MLP and RBf