Reproduce “RELIEF: Reinforcement Learning Empowered Graph Feature Prompt Tuning” 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

Jiapeng Zhu, Zichen Ding, Jianxiang Yu, Jiaqi Tan, Xiang Li, Weining Qian

RELIEF introduces a Reinforcement Learning framework to adaptively select nodes and generate feature prompts for pre-trained Graph Neural Networks, achieving superior performance and data efficiency in few-shot graph and node classification tasks with minimal prompt intrusion.

Key insights

  • Judiciously inserting 'necessary' and 'lightweight' feature prompts to only *certain* graph nodes significantly enhances downstream task performance without disrupting the pre-trained GNN's internal knowledge.
  • Reinforcement Learning is an effective approach to solve the complex combinatorial optimization problem of adaptively determining both the location and content of prompts in graph feature spaces.
  • The use of loss decrease as an instant reward signal effectively guides the RL agent to learn policies that generate powerful, yet non-intrusive, prompts, optimizing for performance while minimizing disturbance.

Summary by alphaXiv.

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