Reconstructing LaTeX Source Files from Generated PDFs - a Neural Network Approach

Deep neural networks (DNN) have been successfully used to solve classification problems for many years. DNN has found variety of applications in recent years beyond classification, including generative tasks. In this paper, we present a neural network architecture suited for generating LaTeX source code which reconstructs a given PDF file. Although a challenging problem to solve, the network performs reasonably well, achieving approximately 70% accuracy rate on validation data.