thanks kt. i was wondering if you had an opinion on which method (NN or a matrix based approach) would be faster to compute / recompute when receiving new datapoints every few seconds?
Well NN is a scalar/vector/matrix approach depending on how you initialize your input node structure. Actually, regardless of what kind of algorithm you use for curve fitting, it all depends on the results you are intending to receive. In general, curve fitting can be dangerous in that results may very well be over fit for your true needs. That's why most people stick with linear regression methods. As I said it all depends on your needs and how well your ability is in applying the algorithm. Another couple pieces of software for learning and applying NNs, etc... that might interest you, are... LNKnet and SNNS
The math guys call it "Nonlinear Optimization" and the most famous algorithms to do it are called Davidon-Fletcher-Powell Levenberg-Marquardt Polytope-Simplex Simulated Annealing Source code for these is in the book Numerical Recipes