This program will address novel neural network models,for distributed stochastic operations. Distributed stochastic operations arise in a number of interesting systems, such as: (1) Multisensor target tracking, (2) Distributed and hierarchical decision making in organization structures, (3) Distributed and cooperative computer data processing, and (4) Global forecasting of economy factors in complex distributed systems. The neural network modeling of such systems is highly important, since it can dictate appropriate operations to be executed and important factors to be extracted and communicated, for effective system operation with simultaneously reduced complexity. Such modeling may also assist in the understanding of the basic elementary features and processes in complex systems as the above. The P.I. considers neural network structures whose objective is either hypothesis testing or parameter estimation. She will test specific approaches for the design and analysis of such networks, with performance criteria matching to the environment the network operates in and its objectives. The P.I. will also investigate new adaptive network training algorithms, with superior robustness and convergence characteristics. These algorithms outperform far the popular Least Mean Squares (LMS) algorithm, using error criteria that are more relevant than the minimum mean square error for the classification of patterns.
