Multi-valued and universal binary neurons: mathematical model, learning, networks, application to image processing and pattern recognition

Conception of universal binary neurons and multivalued neurons with complex-valued weights and their applications to image processing and pattern recognition are considered in this paper. First, efficiency of the “passage” to the complex domain for increasing of the neuron´s functionality is considered. A solution of the XOR problem on the single universal binary neuron is considered. The high speed learning algorithm for the both neurons is developed. Next, neural networks with cellular and random connections based on the considered neurons are proposed. Applications of such networks to image processing (cellular) and image recognition (random) are proposed. The use of multi-valued neurons for time-series extrapolation is also considered