Yilin Wang (王一霖)

Hi, thanks for visiting my website! I am a final-year B.E. student in Computer Science at Xi'an Jiaotong University and an incoming CSE Ph.D. student at Chinese University of Hong Kong, advised by Prof. Xiusi Chen.

I spent summer 2026 at Westlake NLP, interning with Yue Zhang, Dr. Michalis, and Ph.D. student Zhizhang Fu. My earlier work includes NLP applications and mechanistic interpretability.

Email:  yilinw376 [at] gmail [dot] com

CV  /  Google Scholar  /  Twitter  /  GitHub

1) post-training, 2) interpretability, and 3) agents memory

We train language models from scratch in a symbolic environment and find that the failure of two-hop reasoning stems from the asymmetric function the upper layers learns w.r.t. entity representations across different distributions. We then propose a looped structure to fix the problem.

In this work, we propose CSKS (Continuously Steering Knowledge Sensitivity), a simple framework that can steer LLMs' sensitivity to contextual knowledge continuously at a lightweight cost.

• Co-lead of the LUD lab (undergraduate research group @ XJTU), 2026–2027.

Website template from Jon Barron.