Modern machine learning techniques have become indispensable in many fields of astronomy and astrophysics. Here we introduce a specific class methods, invertible neural networks, discuss two applications, the prediction stellar parameters from photometric observations study feedback processes on emission lines.