12112_ood_link_prediction_generaliza

In Appendix A, we introduce more related work that has not been discussed in the main paper. Appendix B, we provide more details in experiments set up and model training. In Appendix D, we show large random and real world graphs have few isomorphic nodes. Ability of GNNs to emulate graph algorithms as graph sizes increase. Our experiments show that the max aggregator, just like the sum aggregators, shows poor OOD performance as test graph sizes increase.