Develoment of a General Purpose Neural Network Simulator (NNSim)

Abstract

Neuron is the basic unit of nervous system. Neural networks are networks of these neurons. In a neural network, many different sets of weights can approximately realize an input-output mapping. The neural network process information through a nonlinear activation function and its sensitivity depends on the weights, which are adapted to improve performance. The process of weight adaptation is treated as learning and learning techniques can be classified as supervised and unsupervised learning. In the present work, an attempt has been made to develop a general purpose Neural Network Simulator (NNSim). Starting from -the biological neuron to the artificial neuron and giving historical background of the field. it has been tried to describe how neural networks can be built form individual nodes. An overview of emerging applications and characteristics of neural networks is also given. As this Simulator (NNSim) is based on the back propagation learning algorithm, the detailed discussion of backpropagation neural network architecture and algorithm is also covered. Two problems. namely Exclusive-OR Logical Function and Computation of Thermal Efficiency of Recovery Unit of paper making process using Neural Network, are also discussed to show the working of the this simulator.