Stable States and Sufficient Conditions for Correct Retrieval in the Bidirectional Associative Memory

Bidirectional associative memories are used in pattern recognition applications and store the given input-output pairs. An important consideration in pattern association applications is the stability and recall accuracy of a given neural network. Some conditions for stability of library vectors using Hamming distances in a bidirectional associative memory are discussed. We also discuss the conditions for correct retrieval of stable states in a bidirectional associative memory. We derive the conditions for correct retrieval in terms of the maximum and minimum of the correlations as well as the number of pattern pairs and the size of the pattern vectors.