Dynamic Compensation of Nonlinear Sensors by a Learning-From-Examples Approach