Machine Learning As A Communications Theory Problem

Machine learning, by far, has been the most impactful research discipline in the last decade. Motivated by the success of machine learning, almost all branches of academia have attempted to apply machine learning algorithms for solving some of their open problems. For example, employing machine learning for improving the performance of communications schemes is one of the main research directions in wireless communications. In this article, we will argue that the communications research community should also do the opposite, i.e., to use the communications\information theory framework for improving machine learning. To this end, we will show that the main machine learning problems can be represented equivalently as communications theory problems. This communications theory representation of the machine learning problems will enable researchers from the fields of communications\information theory to see the machine learning problems as familiar problems and thereby enable the vast body of knowledge and intuition from communications\information theory to be used for tackling the open problems in machine learning.