An Efficient Recall in Diversified Training Samples using Bidirectional Associative Memory

This work attempts to understand the intricacies of the working model of Neural Networks in pattern recognition. The system recognizes the input pattern against the stored ones. It also accepts some decent amount of noise in the input pattern and aims at efficiently recognizing the pattern correctly. The objective of this system is to find how effectively it recognizes characters that are stored as patterns in the system and map the input to the stored pattern, when the input patterns are diversified. And in order to achieve this, the idea of Bidirectional Associative Memory is used. Bidirectional Associative Memory is a two level non linear neural network. One important performance attribute of discrete BAM is the ability to recall the stored pairs particularly in the presence of noise. This is one of the main objective of the system, to recognize patterns in the presence of some permissible noise and study how the system has problems with recalling correct patterns when the training samples are not diversified.