Graph neural networks are intimately related to partial differential equations governing information diffusion on graphs. Thinking of GNNs as PDEs leads to a new broad class of graph ML methods.