Reproduce “Relational Attention: Generalizing Transformers for Graph-Structured Tasks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Cameron Diao, Ricky Loynd
The relational transformer gives entities and directed relations separate vectors, uses both in attention, and updates both during computation. On CLRS-30, its reported class-average test micro-F1 was 66.60% across eight algorithm classes, versus 49.23% for MPNN, leading six classes. Models were tuned over 20 seeds and tested on 64-node graphs after 16-node training. The comparison covers GNNs, not newer graph-transformer methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.