Adaptive Regularization of the Reference Model in an Inverse Problem