Aim 1 (node-centric): learn with labels per node or embed nodes in their neighborhood.
Aim 2 (node-centric): learn with labels per node in >1-neighborhood, make stochastic use of multiple observations per graph.
Aim 3 (graph-centric): learn with labels on the entire graph or embed the entire graph.
Aim 4 (edge support): Support models with non-scalar edge representations