[attachment=12518]
1. INTRODUCTION
1.1 Seminar Context
One of the most popular applications of neural networks is distinguishing patterns into two or more disjoint sets. The neural network approach has been applied to solve pattern classification problems in diverse areas. While the predictive accuracy obtained by neural networks is usually satisfactory, it is often said that a neural network is practically a black box. Even for a network with only a single hidden layer, it is generally impossible to explain why a certain pattern is classified as a member of one set and another pattern as a member of another set, due to the unimaginable complexity of the network.
In many applications, however, it is desirable to have rules that explicitly state under what conditions a pattern is classified as a member of one set or another. Several approaches have been developed for extracting rules from a trained neural network.
The process of rule extraction from a trained network can be made much easier if the complexity of the network has first been reduced. The pruning process attempts to eliminate as many connections as possible from the network, while at the same time maintaining the pre specified accuracy. It is expected that small number of connections will result in more concise rules. Relevant and irrelevant attributes of input data are distinguished during the pruning process.
Now, rules have to be extracted from the pruned network. This process involves two basic steps. First, clustering the activation values of the hidden units into a smaller number of clusters and second, splitting the hidden units into a number of units.
Generating a compact set of classification rules is essential for effective use of the rules generated. For this purpose, Cluster Validation is indispensable. These clusters obtained are validated class wise. Since it is essential to avoid forming combinations with insignificant clusters cluster validation becomes an essential part of rule generation process.
The rule generation scheme employed in this methodology can be shown as below
2. LITERATURE SURVEY
2.1 Introduction to Artificial Neural Networks
2.1.1 Description and Definition
An artificial neural network can be defined as a system composed of many simple processing elements operating in parallel. Each element has an input/output (I/O) characteristic and implements a local computation or function. The output of any unit is determined by its I/O characteristic, its interconnection to other units, and external inputs. Although it is possible to hand craft the network, the network usually develops an overall functionality through one or more forms of training.
A neural network is a massively parallel distributed processor that has a natural propensity for storing experimental knowledge and making it available for use. It resembles the brain in two respects.
1. Knowledge is acquired by the network through a learning process.
2. Interneuron connection strengths known as synaptic weights are used to store the knowledge.
Distributed computation has the advantages of reliability, fault tolerance, high throughput and cooperative computing, but generates the problems of locality of information and the choice of interconnection topology. The features of distributed processing, adaptation and nonlinearity, are the hallmark of biological information systems, and hence of neural networks.
Learning typically occurs by example through training, or exposure to a set of input/output data where the training algorithm iteratively adjusts the connection weights. These connection weights store the knowledge to solve specific problems. This procedure is referred to as the learning algorithm.
Neural networks with their remarkable ability to derive meaning from complicated or imprecise data can be used to extract patterns and detect trends that are too complex to be noticed by either humans or other computer techniques. A trained neural network can be thought of as expert in the category of information it has been given to analyze. Advantages of neural networks include-:
1. Adaptive learning: An ability to learn how to do tasks on the data given for training or initial experience.
2. Self Organization: An ANN can create its own organization or representation of the information it receives during learning time.
3. Real Time Operation: ANN computations may be carried out in parallel, and special hardware devices are being designed and manufactured which take advantage of this capability.
4. Fault Tolerance: Partial destruction of a network leads to the corresponding degradation of performance. However, some network capabilities may be retained even with major network damage.
2.1.2 Applications
Neural networks are used for the problems that do not have an algorithmic solution or for which an algorithmic solution is too complex, and are often well suited to problems that people are good at solving, but for which traditional methods are not. Their applications are limitless but fall into a few simple categories.
1. Prediction
2. Classification
3. Data Association
4. Data Conceptualization
5. Data Filtering