High-Speed Adaptive Optimal Associative Memory

Cyber Dynamics, Inc. (CDI), proposes to develop a software implementation of the theoretically fastest adaptive process for Optimal Associative Memory (OAM). This algorithm provides orders of magnitude improvement over all conventional neural network paradigms in the areas of response recall accuracy, information storage capacity and density, immunity to noise, generalization characteristics, and effective rates of processing. There is no practical limit to the number or complexity of Stimulus-Response patterns which may be enfolded onto a single correlation matrix. Response error during training rapidly converges to zero with few iterations required to attain near perfect recall. Nonlinearity within pattern associations is not a limiting factor. Stimulus patterns are transformed in parallel fashion through all patterns previously encoded into the matrix. Response values provide not only the desired information but also a measure of confidence in that information. Pattern recognition performance and speed far superior to that of the fastest supercomputer doing linear search may be achieved with ordinary personal computers using OAM.