Reproduce “A Foundation Model for Zero-shot Logical Query Reasoning” 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
Mikhail Galkin, Jincheng Zhou, Bruno Ribeiro, Jian Tang, Zhaocheng Zhu
Researchers developed ULTRAQUERY, a foundation model enabling zero-shot complex logical query answering on any knowledge graph, including those with unseen entities and relations. This model achieved up to 70% relative improvement over baselines for negation queries and approximately 50% for other query types, particularly on newly introduced inductive benchmarks.
Key insights
- Decoupling complex logical query execution into an inductive, vocabulary-independent relation projection operator and non-parametric logical operations enables zero-shot generalization across KGs.
- Fine-tuning the inductive projection operator on complex logical queries is essential to address the distribution shift caused by multi-source propagation in multi-hop reasoning.
- A model trained on a single or a few transductive datasets can generalize effectively to a wide range of unseen KGs with distinct entities and relations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.