FMSTS accepted at CPAIOR 2020

Our paper Reinforcement Learning for Variable Selection in a Branch and Bound Algorithm has been accepted at CPAIOR 2020!

This work introduces FMSTS, a novel Reinforcement Learning approach specifically designed for branching in Branch and Bound solvers. By leveraging patterns in real-world instances, FMSTS learns from scratch a branching strategy optimised for a given problem. The method is grounded in a consistency between a local value function and a global metric of interest, and features a new neural network architecture. To our knowledge, this is the first time RL has been used to fully optimise the branching strategy.

Authors: Marc Etheve, Zacharie Alès, Côme Bissuel, Olivier Juan, Safia Kedad-Sidhoum

Links: Project page · arXiv · HAL · Springer