I am an undergraduate student at Renmin University of China. My research interests include LLM reasoning, Transformer architecture optimization, post-training, and multimodal AI safety.

Broadly, I am interested in how foundation models reason and how to make that process more effective and efficient. I also care about the safety and reliability of multimodal systems.

I am fortunate to work with Prof. Ruihua Song, Prof. Suyun Zhao, and Prof. Wenxuan Wang at Renmin University of China.

Selected Publications

arXiv 2026

Can Computation from Earlier Problems Help LLMs Solve New Ones?

Jipei He, Wenhui Tan, Xiaoyi Yu, Enver Sangineto, Fiorenzo Parascandolo, Rita Cucchiara, Ruihua Song

arXiv, 2026

We study whether computation retained from earlier problems can help later reasoning, and introduce STAIR, a lightweight mechanism for reusing historical attention states.

arXiv PDF HTML

Research Interests

• LLM Reasoning — reasoning mechanisms and inference-time computation.
• Transformer Architecture Optimization — efficient architectural design for large language and multimodal models.
• Post-Training — methods for improving reasoning and generalization after pretraining.
• Multimodal AI Safety — safety and robustness across modalities.