Or write a Genetic Algorithm (using AI::Genetic for example). Use a variety of functions as genes from which to build the genome of each 'individual'. The fitness test could use least squares.
in reply to Re: Estimating continuous functions
in thread Estimating continuous functions
Problem with Neural Nets is they tend to be black boxes, though they might approximate your function they can't easily tell you what it is. Unless you implement a *really* big one that can parse and generate natural language as well as your function ;)