胡雨韦

Yuwei Hu

I am currently a junior undergraduate student of Gaoling School of Artificial Intelligence, Renmin University of China. I am honored to be advised by Prof. Zhewei Wei.

Email / Github

Research Interests

My past research experience is mainly in the fileds of graph neural networks, including graph transformer, graph lottery tickets.

Publications

Intruding with Words: Towards Understanding Graph Injection Attacks at the Text Level

Runlin Lei, Yuwei Hu, Yuchen Ren, and Zhewei Wei.
To appear in NeurIPS 2024, arXiv preprint arXiv:2405.16405, 2024.

• We are the first to unveil the dynamics of text-level graph adversarial attacks on TAGs, especially for GIAs.
• We propose three effective text-level GIAs and demonstrate the trade-off between attack performance and text interpretability.
• We reflect on the challenges of graph adversarial attacks at the text level.

Scalable and Accurate Graph Reasoning with LLM-based Multi-agents

Yuwei Hu, Runlin Lei, Xinyi Huang, Zhewei Wei, and Yongchao Liu.
arXiv preprint arXiv:2410.05130, 2024.

• We propose the first LLM-based multi-agents fine-tuning-free framework for graph reasoning.
• We expand the scale of graph reasoning tasks handled by LLMs from 100 nodes to 1,000 nodes.
• Our framework demonstrates the ability to tackle real-world applications like webpage importance analysis.

Selected Honors and Awards

None, none, 2023