Neural Networks

Introduction

Work on neural networks, usually referred to as "neural networks," has been motivated right from NN inception by the recognition that the human brain computes in an entirely different way from the conventional digital computer. In its most usual form, a neural network is a machine that is designed to model the way whether the brain performs a particular task or function of interest; the network is generally implemented by utilizing electronic components or is simulated in software on a digital computer. We may present the following definition of neural network seen as an adaptive machine:

A neural network is a massively or particularly parallel distributed processor made up of easy processing units such has a natural propensity for storing experiential knowledge and making it available for employ. This resembles the brain in two respects as:

• Knowledge is acquired by the network from its environment throughout a learning process.
• Interneuron connection strengths, termed as synaptic weights, are utilized to store the acquired knowledge.

Neural networks are a class of computing systems that employ a highly parallel architecture to efficiently perform pattern recall, classification, and prediction tasks. These networks are loosely modeled after human networks of neurons in the brain and nervous systems. The networks consist of large numbers of simple processing units that are characterized by a state of activation, which is a function of inputs to the units. All units calculate three functions as:

• Propagation Function Net_i : Generally the propagation function is calculated as the weighted sum of the outputs of all other units connected to unit i :

                                         ....................Eqn (1)

 Whereas, W_ij is the strength of the connection in between units i and j, and the output level of unit j is O_j.

• Activation Function A_i : A function of Net_i and some time.

Commonly utilized functions are the linear, threshold, and logistic functions.

• Output Function O_i : A function of A_i, O_i is often easy set equal to the activation, A_i.

The connections may be excitatory or inhibitory. Inhibitory connections tend to decrease the activation of units to which they are connected, while excitatory connections will tend to raise or increase the activation.

Neural network operate at two various levels of time:

-          Short-term response to environmental inputs, and

-          Long-term changes in connection weight that encode knowledge and change how the network reacts to its environment

Various neural network architectures have been developed. They differ in the categories of propagation and activation functions utilized, how units are interconnected, and how learning is implemented. The kind of paradigm employed depends on the characteristics of the task to be performed. A most distinction among networks is whether the system will be employed for recall or says recognition, classification, or prediction. Recall systems are employed for noise filtering and pattern completion and also called CAM or content addressable memories. Illustrations are Anderson Brain State in a box or BSB model and Hopfield network.

Prediction systems can be utilized to estimate the behavior of complex systems. An illustration is the work of Rangwala and Dornfield on prediction of machining parameters. Classification networks makes mappings of input patterns into categories, presented by characteristic output patterns.

The utilization of neural networks offers the following helpful properties and capabilities as: Non-linearity, Adaptivity, output-input Mapping, Evidential Response, Fault Tolerance, Contextual Information, Uniformity of Design and Analysis, Neurobiological Analogy.