Reproduce “RelICL: Training-free Relational Learning with Tabular Foundation Models” 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
Simon Forbat, Rainer Gemulla
RelICL predicts labels for entities using information distributed across linked tables. It propagates keys, encodes whole rows with a tabular foundation model, and pools child-row representations into parent rows before prediction. On a constructed interaction task, RelICL with TabICL v2 reached 99.3 ± 0.7 AUROC, while RDBLearn scored 46.5 ± 3.3; this does not establish superiority on ordinary benchmarks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.