1 paper across 1 session
We treat substructures as graph tokens to enable the scale transformer pre-training on graphs.