Title: Learning Combinatorial Optimization on Graphs: A Survey with Applications to Networking

Abstract: Existing approaches to solving combinatorial optimization problems on graphs suffer from the need to engineer each problem algorithmically, with practical problems recurring in many instances. The practical side of theoretical computer science, such as computational complexity then needs to be addressed. Relevant developments in machine learning research on graphs is surveyed, for this purpose. We organize and compare the structures involved with learning to solve combinatorial optimization problems, with a special eye on the telecommunications domain and its continuous development of live and research networks.