DropGNN: Random Dropouts Increase the Expressiveness of Graph Neural Networks

This paper studies Dropout Graph Neural Networks (DropGNNs), a new approach\nthat aims to overcome the limitations of standard GNN frameworks. In DropGNNs,\nwe execute multiple runs of a GNN on the input graph, with some of the nodes\nrandomly and independently dropped in each of these runs. Then, we combine the\nresults of these runs to obtain the final result. We prove that DropGNNs can\ndistinguish various graph neighborhoods that cannot be separated by message\npassing GNNs. We derive theoretical bounds for the number of runs required to\nensure a reliable distribution of dropouts, and we prove several properties\nregarding the expressive capabilities and limits of DropGNNs. We experimentally\nvalidate our theoretical findings on expressiveness. Furthermore, we show that\nDropGNNs perform competitively on established GNN benchmarks.\n