NeurIPS 2020

Hybrid Models for Learning to Branch


Meta Review

The paper proposes a CPU version based on a recent machine learning algorithm for selecting branching variables in branch and bound for ILPs. The hybrid method is based on a GNN together with a neural net to form a prediction. The reviewers agree on the practical benefits this brings and agree on the work being a solid experimental and engineering work. Reviewers criticize that the work is incremental and the ideas used are found in previous work. One reviewer questions the practical significance of the work for the future. The overall recommendation for the paper is accept, since the paper overall represents an interesting practical contribution,