Abstract:

State-of-the-art Mixed Integer Linear Program (MILP) solvers combine systematic tree search with a plethora of hard-coded heuristics, such as the branching rule. The idea of learning branching strategies from data has received increasing attention recently, and promising results have been obtained by learning fast approximations of the strong branching rule. In this talk, we will discuss a different paradigm: learning to branch from scratch via Reinforcement Learning.