How does learning/training work on a GNN? Is there any concept of forward and backward propagation in a graph that is not a DAG?
(Also, some pictures would help.)
