Index Code Construction via Deep Matrix Factorization

In this paper, we consider the problem of on-demand source coding for collaborative data dissemination in vehicular ad-hoc networks (VANETs). Specifically, we address the problem of index code construction for a big-data network. In this setup, users request a certain subset of data from a central server, while possessing some side information. The central server broadcasts a compressed message to meet all the users' demands. We consider a practical and generalized scenario where the side information and requested data packets of a user may be coded. Constructing index codes for such a setup is known to be an NP-hard problem. For this scenario, we propose a deep learning based approach to construct index codes through low-rank matrix completion. The proposed deep neural network performs matrix factorization to obtain a low-rank index code. We show that the proposed deep learning based approach to construct index codes is not only computationally feasible for networks with large data packets, but also provides better performance compared to the alternating minimization based approach for index code construction.