GNNAutoScale: Scalable and Expressive Graph Neural Networks
              via Historical Embeddings

             Matthias Fey 1 Jan Eric Lenssen 1 Frank Weichert 1 Jure Leskovec 2

             Abstract                 While the full-gradient in a GNN is straightforward to com-
   We present GNNAutoScale (GAS), a framework          pute, assuming one has access to all hidden node embed-
   for scaling arbitrary message-passing GNNs to         dings in all layers, this is not feasible in large-s ...