Towards learning integral strategy of branch and bound

Abstract

Branch and bound is the preferred algorithm used for solving MILP problems. It involves two fundamental strategies that are node selection strategy and branching strategy. Whereas the learning literature has been focused in dealing with just one strategy on the same time, we design a two-in-one strategy of branch and bound algorithm regarding the fact that are intuitively dependent. To do so, we apply the well-known SVM algorithm to the well-known set of problems MIPLIP.