On the representation by bivariate ridge functions (Q2058579)

Ridge functions are an important tool in approximation theory because they form the basics of neural network and learning research. In this paper, bivariate ridge functions are considered. One of the main questions is of course whether (continuous, say) functions of any type may be expanded as linear combinations of ridge functions. In this article, the authors take the point of view that this may be carried out in a two-step process. First, a provided approximant is assumed to be written as a linear combination of \textit{any} ridge functions. Then they show that it can be transformed into an expansion of a certain (user-)provided ridge function.