We shall discuss the problem of constructing index codes that maximize a data network's throughput and minimize power consumption. In this setup, multiple users, with some side information, demand certain subsets of data from a central node. This scenario is the basic model of the data sharing network in various cyber-physical systems and connected networks. The goal of the index code is to minimize the required transmission rate while ensuring that the users can recover the demanded data using an encoded broadcast message from the server. Thus, minimizing the power expended for communication while maximizing throughput. We shall discuss a generalized version of the index coding problem, where both the side information and user-demanded data can be coded. The problem of index code construction, for a given set of side information and users’ demands, can be modeled as a matrix completion problem. Traditional machine learning algorithms used for matrix completion do not take advantage of the inherent structure in the index codes to construct efficient index codes. We shall discuss novel techniques proposed by us for low-rank factorization of structured matrices. We shall see that the proposed machine learning methods for structured matrix completion construct efficient index codes compared to the traditional methods known in the literature so far.
Dr. Lakshmi N. Theagarajan obtained his MS and PhD from the Indian Institute of Science in 2015. In his doctoral research, he developed novel low-complexity probabilistic graphical model based inference algorithms that made massive MIMO signal processing methods practically realizable. In 2016-2017, he was a postdoctoral researcher at the Sensor Fusion Lab at Syracuse University, where he developed novel online statistical learning algorithms for distributed detection and estimation problems. Currently, he is an assistant professor at the Indian Institute of Technology Palakkad and a visiting assistant professor at McMaster University. He has also worked in Cisco Systems R&D and National Instruments R&D in the past. His areas of research include statistical learning and inference in large-scale wireless networks, distributed machine learning (federated learning), sparse signal processing, design of optimal modulation and coding schemes, visible light communication, information and coding theory.