Abstract
In practice, algorithms are often run repeatedly on similar problems from a particular application domain. For example, shipping companies solve similar routing problems day after day, and although demand changes daily, the underlying problem structure (such as the road network) remains the same. A growing body of research has studied how to use data about the application domain together with machine learning to optimize an algorithm so that it has particularly strong performance on problems from that application. In this talk, I will give an overview of research that uses machine learning in the context of algorithm design. In particular, I will focus on research that provides theoretical guarantees on the performance of the resulting algorithm.