Can we learn combinatorial algorithms on billion-scale graphs? Our algorithm GCOMB shows you can! GCOMB is possibly the fastest algorithm for Influence Maximization. This work will appear in NeurIPS 2020.