BIOMASSCOMP: Artificial Neural Networks and Neurocomputers

摘要

BIOMASSCOMP is a project whose objective is to define and develop methods for automating the process of 'reverse engineering' the brain for application to the development of intelligent sensors and controllers for avionic and other systems. What we have done in this project is to quantify and apply concepts that many neural network and cognitive science researchers have tacitly and qualitatively assumed to be work in self-organizing systems. During this Phase I SBIR project, we have defined, developed, and implemented an entropy-based scalar measure, DMORPH, of the common structure between two systems, as evidenced by measurement of signals from the two systems. By design, DMORPH reflects only the crosscorrelations between systems and not the intracorrelations within the separate systems. DMORPH was applied to the input and output signals from various artificial neural network architectures to attempt to determine which networks, and which parameter settings within each, induced the greatest structural similarity between input and output signals after learning had taken place. This research applies to the development and testing of real time autonomous learning systems suitable for application to problems of avionics sensor fusion, adaptive sensor processing, and intelligent resource management. (AW)