Todd, P.M. (1991). Neural networks for applications in the arts. In M. Scott (Ed.), Proceedings of the Eleventh Annual Symposium on Small Computers in the Arts (pp. 3-8). Philadelphia, PA: Small Computers in the Arts Network, Inc.
Abstract
A new approach to computer art applications is provided by the rapidly expanding field of neural networks. Rather than operate in the traditional style of preprogrammed rule-following systems, neural networks have the power to *learn* to produce specific types of outputs from specific inputs, based on examples they are taught. Thus, instead of having to specify *how* to create a certain artwork, the artist can instead teach a network *examples* of the desired output, and have the network generate new instances in that style. We need not specify the steps involved in creating a Rodin sculpture--we can just collect instances of the sorts of sculpture we'd like to get and use those to train a network. This sort of thing is the promise of neural network applications in the arts, stated in a very provocative and exaggerated manner. But we are beginning to realize this promise in limited domains, and the space for further applications and explorations of these techniques is wide open. In this paper, I will briefly describe neural networks, some of what they can and can't do, some of what they have done already in the arts, primarily music, and some of what they can be applied to in other areas.
Back to main page.