Summary form only given, as follows. Several features found innneural networks make them desirable for music composition. Thesenfeatures include the ability to learn complex relationships fromnexamples and to make reasonable generalizations in situations for whichnthey have not been trained. Training consists of exposing the network tonexamples of music and adjusting its internal representation to match theninput examples. It is this internalization of the music which representsnmany aspects and relationships that humans perceive both consciously andnunconsciously. The method used to present the music to the network willntherefore largely affect what features the network will extract. Severalnstrategies involved in creating music using neural networks have beennexamined, and work involving multiple input training which mimics thenfunctions of the right and left hemispheres of the brain has beenncarried out
展开▼
