It seems like GraphLab would be better suited for ML algorithms like recursive neural nets that require back propagation since they're all about sending messages between nodes to determine convergence.

Is it even possible to implement back prop with Ligra?