Reproduce “GRIL: Knowledge Graph Retrieval-Integrated Learning with Large Language 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

Jialin Chen, Houyu Zhang, Seongjun Yun, Alejandro Mottini, Rex Ying, Xiang Song, Vassilis N. Ioannidis, Zheng Li and 1 more

GRIL introduces an end-to-end framework for Knowledge Graph Question Answering that integrates graph retrieval and Large Language Model reasoning. The system employs an attention-based retriever and uses LLM output logits as implicit feedback for joint optimization, enabling state-of-the-art performance on KGQA benchmarks and robust operation in open-domain settings with smaller language models.

Key insights

  • Directly coupling the graph retriever and LLM reasoner through joint optimization, using implicit LLM feedback, significantly improves the usefulness of retrieved knowledge for complex reasoning tasks.
  • An adaptive graph retriever with growing, pruning, and a complexity assessment module can efficiently navigate large KGs and reduce computational overhead.
  • GRIL enables smaller, more efficient LLMs to achieve competitive or superior performance compared to larger, proprietary models by intelligently leveraging structured knowledge.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.