Incremental evolution of weight modifiers for neural networks in cooperative behavior generation

The purpose of this research is to develop methodology for generating large and complex neural networks. Especially, in this case, we adopt the methods to the organizational task in a multi-agent system. We propose the incremental evolution of weight modifiers for generating neural networks based on the GA. In this method, a chromosome contains only the instructions for making and modifying the weights in the base model of the neural network. In this generating process, because the GA holds only the instructions, the search space will be smaller than that of direct encoding methods. Concerning the architecture of the base model, we only determine the maximum number of neuron nodes. As an experiment for the proposed methods, we apply them to the organizational behavior generation in a multi-agent system. Throughout the experiments, the effectiveness of our proposed methods is shown