Recipe for a General, Powerful, Scalable Graph Transformer

NeurIPS, 2022
← Back to References

Bibliographic Information

TitleRecipe for a General, Powerful, Scalable Graph Transformer
AuthorsLadislav Rampášek, Mikhail Galkin, Vijay Prakash Dwivedi, Anh Tuan Luu, Guy Wolf, Dominique Beaini
VenueNeurIPS
Year2022
Linkhttps://arxiv.org/abs/2205.12454

Summary

Introduces GraphGPS, a modular framework combining local MPNN with global attention and flexible positional/structural encodings. Demonstrates that the combination of local message passing + global attention + good PE is a powerful recipe. STRATA's reachability profiles offer a potential alternative PE for directed graphs.