*** 9704190 Conrad Artificial neuromolecular (ANM) architectures are brain-like computer designs in which the input-output capabilities of the neuronal units are controlled by internal dynamics that serve to fuse signals in space and time. These internal (or intra-neuronal) dynamics are motivated by molecular process believed to be operative in real neurons. The pattern processing and control capabilities of the global architecture this draw on neuronal as well as network level dynamics. Evolutionary algorithms orchestrate this repertoire into groupings suitable for coherent perception-action tasks, including tasks involving complex sequences of actions. The system is being applied to Chinese character differentiation and recognition/effector control behavior in a maze-like environment. The objective of this project is to develop a methodology for designing networks to increase the effectiveness of evolutionary computing as an adaptation technique. The AM architecture is use as the testbed. Four submodels with different types of dynamics have bee developed for this purpose. The first is an abstract (dual dynamics), network model that consist of pattern recognizing components subject to weak collective interactions. The second, the cytomatrix neuron, is a softened cellular automaton. the third is a coupled oscillator model motivated by the neuron cytoskeleton. the fourth is a general simulation system that can be used to model motivated by the neuronal cytoskeleton. The fourth is a general simulation system that can be used to model a wide variety of intracellular dynamic processes involving the co-action of kinetic and structural processes. These different submodels con be studied in their own right and also can be embedded in the neural units of ANM system. The comparative experimental study of these models complement theoretical analysis and provided guidelines for adaptive surface engineering pertinent to a wide variety of neural architectures. ***
