Solution to the problem of training a multi-agent neural network by means of a multi-chromosome genetic algorithm

This paper proposed a concept of the multi-agent neural network, where neurons are agents of the multi-agent system. Each agent has a decision-making system that rests upon the system of production rules. A multi-chromosome genetic algorithm was developed. This algorithm allows to train a multi-agent neural network simultaneously by several parameter classes of the target function. Efficiency of the proposed training method was illustrated by solving the task of searching behavior¹.