Use of Conditional Generative Variational Autoencoder Networks to Improve Representativity of Data in Optical Spectroscopy

In this study, the solution of the inverse problem of spectroscopy of water-ethanol liquid solutions by neural network models is considered. The process of training a neural network requires a large number of patterns, which cannot be obtained by laboratory measurements. In this paper, we demonstrate the possibility of generating an additional array of patterns using a conditional variational autoencoder. The generated patterns have a form similar to real spectra, and they are used to train the neural network for classification, along with the original patterns. As a result of applying this approach, it was possible to improve the quality of solving the inverse problem on real patterns that were not used in the training process.