Multidimensional SVM to include the samples of the derivatives in the reconstruction of a function

In this paper we propose a multidimensional regression estimation algorithm for estimating a function from its first derivatives. The proposed method is extended to introduce the information about the function itself and higher order derivatives. The proposed algorithm is able to exploit the dependency between the output variables to provide a better estimation of the function and it guarantees that the estimated derivatives belong to the same function. The method has been validated by synthetic test functions and it has been used to model a MESFET transistor including intermodulation distortion characterization, where the approximation of the derivatives of the characteristic function is mandatory.