Reproduce “S-GAP: Stealthy Graph Attack Primitives for Graph 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

Wenda Li, Tongya Zheng, Shunyu Liu, Hanyang Yuan, Kaixuan Chen, Yu Wang, Yuchen Ying, Haoze Jiang and 3 more

S-GAP coordinates text and edge edits in text-attributed graph attacks through a shared Graph Attack Primitive (GAP) semantic plan, iteratively revised with surrogate-model feedback. In transfer attacks, it has the highest attack success rate in all 24 tested dataset–model–setting combinations, including 44.95% versus 38.28% on Cora against LLaGA in graph modification; this finding is limited to tested settings.

Summary by alphaXiv.

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