Inheritance of information in multi-layer sigma-pi neural networks

This article shows that prior knowledge may be incorporated into a neural network by using the knowledge in a trained net to prescribe the weights for a new multi-layer artificial neural network. In this article, reuse of information may be viewed as inheritance of knowledge. The inherited knowledge, in this case, takes the form of weights which are inherited from a previously trained network. The information reuse can either be cast in a geometric guise or into an algebraic guise. The purpose of this paper is to address problems which previous research [R.S Neville, 1998] has not solved.