Using method integration transfer learning for neural network solution of an inverse problem in optical spectroscopy