Ontogenic Neural Networks for Avionics Applications

The study of neural networks which possess the capability to learn and grow with little or no supervision will be explored during this research effort. These networks exhibit ontogenic behavior and are termed ontogenic neural networks. This reasearch effort explores the use of hybrid neural network structures which are capable of self-organization, feature discovery and self generation in a composite architecture which is capable of solving pattern recognition tasks. These networks will be examined for suitability in automatic target recognition, threat assessment, route planning or another problem selected by the Air Force and Accurate Automation. The ontogenic neural network developed in this effort will be thoroughly examined for learning rate, generalization capability, classification accuracy, representation of higher order relationships, self-organization ability and hardware implementation.