Evolutionary approach to creating a neural network model of collective decisions of intellectual tasks

The article examines the possibility of using an evolutionary approach to improve implementation of neural networks and self-learning mechanisms for solving problems based on multi-agent representation of knowledge. The collective use of artificial neural networks as a neural network of agents can further parallelize and distribute between local agents the processes of solving complex intellectual tasks. The algorithms of integrated evolutionary search of the weights to solve a number of learning objectives are described. We propose a genetic algorithm, generating neural network model of optimal topology. In the present genetic algorithm each individual represents a separate neural network, and the population is considered as an evolving multi-agent system in which the strategy of behavior of each agent is determined by its corresponding neural network.