Yihan Wang

Hi, I am a final-year Ph.D. candidate in Computer Science at UCLA, working with Prof. Cho-Jui Hsieh. I completed my B.Eng. degree at Tsinghua University in June 2020.

My research interests focus on the robustness and generalization of machine learning models, with a recent emphasis on large language models. I work on projects that both interest me and benefit society. I was supported by an Amazon Fellowship.

My research specifically includes:

• Identifying and understanding limitations and potential risks in language model fine-tuning, including:
  • Limitations of prompt tuning
  • Two-stage fine-tuning for better generalization
  • Loss of context awareness
• Building robust machine learning models against adversarial attacks:
  • Jailbreaking defense via Backtranslation, llm-jailbreaking-defense
  • Red-teaming LM detectors
  • Fast IBP

I’ve also worked on formal verification of neural networks/machine learning models in my early PhD years.

For anyone interested in my research: Please feel free to email me if you are interested in a discussion on research or potential collaborations.