Reproduce “MERIT: Scaling KG Foundation Models with Relation Transformers” 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
Cosimo Gregucci, Daniel Hernandez, Antonio Vergari, Steffen Staab
MERIT performs zero-shot knowledge-graph link prediction on graphs with unseen relation names by letting all relations attend to one another, with motifs as a learned bias rather than a constraint. On 54 benchmarks, its mean MRR is 0.386, versus 0.384 for TRIX and FLOCK, though it trails them on SQUA. Inference is 0.55 hours versus ULTRA’s 0.58.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.