Parallel architectures and methods of self-organization and complexity minimization for polynomial neural networks (PNNs) designed for image recognition are considered. Estimates of the degree of parallelism in the process of decision making by PNNs are obtained. It is shown that parallelism can be considerably enhanced when complex pattern recognition and image analysis problems are solved collectively (on the multiagent basis). The enhancement of parallelism is then achieved by breaking the global problem down into several local problems whose solution is distributed between self-organizing PNNs considered as neural network agents.
展开▼