Evolution of hierarchical neural networks for time-dependent cognitive processes: key recognition for musical compositions

This paper presents the results of using the GENDALC GANN system to evolve neural network topologies for music perception. The results obtained are not only better than those for other typically used neural network topologies, but also better than for neural networks that incorporate music theory knowledge. Because the data and task used in these experiments include hierarchical time dependent processing, these results demonstrate GENDALC's ability to evolve good solutions for cognitive tasks, even while using approaches potentially different from those used by humans.