Abstract

In a Bayesian framework, we give a principled account of how domainspecific prior knowledge such as imperfect analytic domain theories can be optimally incorporated into networks of locally-tuned units: by choosing a specific architecture and by applying a specific training regimen. Our method proved successful in overcoming the data deficiency problem in a large-scale application to devise a neural control for a hot line rolling mill. It achieves in this application significantly higher accuracy than optimally-tuned standard algorithms such as sigmoidal backpropagation, and outperforms the state-of-the-art solution. 1 INTRODUCTION Learning in connectionist networks typically requires many training examples and relies more or less explicitly on some kind of syntactic preference bias such as "minimal architecture" (Rumelhart, 1988; Le Cun et al., 1990; Weigend, 1991; inter alia) or a smoothness constraint operator (Poggio et al., 1990), but does not make use of explicit representations...

Citations