Neural Network

Artificial neural networks are developed with the aim of brain modeling to retrieve  information. Artificial neurons are inaccessible when compared to biological neurons. Neural network development finds its own applications in the field of computational molecular biology, analysis of protein or genetic material, in recognizing patterns and machine learning tools.  It involves graphical representations that are linked by network units. It looks like a weight directed graph or architecture. Neural network helps in proper assembling of sequence in a hierarchical fashion. It helps in easy prediction of parameters like profit or gains, delays or loss and time constant. Parameters are determined by number of layers that are present, number of units in each layer and also network connections  between each layer.

Types of weight directed graph

The architectures are distinguished in to following types.

• Recurrent architecture
• Feed forward architecture
• Layered architecture

Recurrent architecture: It includes direct loops.

Feed forward architecture: Architecture without direct loops. They are not layered necessarily. Most of the neural networks are feed forward architectures.

Layered architecture: It is partitioned as various units of layers. There are two types of units, namely hidden units and visible units. For instance, input and output units are visible units and are represented as input layer and output layer. Input units are used to code the sequential data and output unit represents the typical representation of functional features. Units which are not visible are referred as hidden units. Layer number is referred as depth of network.

Problems of neural network

Neural network helps in easy sort out of regression and recognition problems. In case of regression problem, main objective is to occupy the available surface area. However, recognition problem or classification problem help in easy sort out of input layer in to the number of classes and are discrete processes.