We propose to use neural networks to learn and replace Monte Carlo (MC) simulations. Our neural networks are not only convenient to use but also are shown to be extremely time-saving compared to MC simulations with comparable accuracy. Furthermore, we employ the machine learning method called transfer learning to perform calibration between MC simulated diffuse reflectance and that measured by Diffuse Reflectance Spectroscopy (DRS). The transfer learning model is able to predict DRS measured diffuse reflectance spectrum by training the model using a small amount of DRS data.