Choosing Algorithm Parameters Strategically

In this paper we examine the potential use of articial intelligence in choosing values for free parameters in the software implementation of algorithms. As a particular example we examine a branch-and-cut procedure for solving integer programs and an implementation of that algorithm called MIPO. We examine a particular parameter in that procedure called the skip factor whose value decides how often cuts are added to the problem description. We demonstrate how carefully choosing the value for this parameter using neural networks results in a decrease in solution time compared to using a fixed value or a value suggested by MIPO