Combinatorial optimization using machine learning

Abstract

Combinatorial optimization is a subfield of mathematical optimization that contains several hard problems with numerous real-life applications. The traditional way of solving combinatorial optimization problems relies on decisions taken based on expert knowledge and expert-designed heuristics. In recent years, a promising research line has brought an alternative way to light. This way is to automate decision- making for combinatorial optimization using machine learning. In this thesis, we provide general information on this research line but focus more on direct ways of leveraging machine learning to solve combinatorial optimization problems. Moreover, we create a reinforcement learning framework that learns a greedy constructive heuristic for the following graph combinatorial optimization problems: minimum vertex cover, maximum independent set and travelling salesman.