Sketch-GNN: Scalable Graph Neural Networks with Sublinear Training Complexity We present a sketch-based GNN training technique with sublinear training time and memory complexities with respect to graph size.