**TODO**: summarize from <https://blog.derwen.ai/graph-levels-of-detail-ea4226abba55>

Graph topological transform approaches so far (e.g., `lee2023ingram`) have focused on using relation affinities to train _representation learning_ models. this may be another example of using deep learning as a mêlée weapon. instead, 

results computed from _graph of relations_ analysis naturally feed into _statistical relational learning_ approaches such as _probabilistic soft logic_, to develop rule sets and ground truth for training SRE models.

TODO: survey/compare topological decomposition of graphs, then using statistics to determine how to reconstruct probabilistically => for recomposition of generate graph elements (not simple nodes, edges)