Density estimation using a generalized neuron

Abstract

Neural Networks have been shown to be useful took for density estimation. However, the training of neural network structures is time consuming and requires fast processors for practical applications. A new method with a Generalized Neuron (GN) for density estimation is presented in this paper. The GN is trained with the particle swarm optimization algorithm which is known to have fast convergence than the standard backpropagation algorithm. Results are presented to show that the GN can estimate the density functions for distribution functions with different means and variances. This density estimation method can also be applied to the multi-sensor data fusion process.