I don't know if this should be in the math forum or the computer science one, but I think it is more relevant in the math section.
I would like to approximate nonlinear functions typically used in neural networks on a given interval. However, the Taylor polynomial approximations are in a neighborhood of a point, leading to overflows in my computations. What's more, I don't think Taylor polynomials are the best ones to used, due to the degree required to fit "well" the function to approximate.
Do you have other algorithms than Chebyshev / Taylor polynomials to approximate functions, that might be faster / more accurate to fit a given function in an interval [a;b]?
Thank you.