Yixin He

Short Bio

yixin.jpg

🏫 Viterbi School of Engineering, University of Southern California

📮 yixinhe@usc.edu

Hi! 👋 I’m Yixin He, a second-year Master’s student in Computer Science (CSAI) at 🏫 USC. I’m currently working remotely as a Research Assistant with Dr. Simin Chen and Prof. Baishakhi Ray from Columbia University. This year, I’m applying to Ph.D. programs to further pursue my research passion ✨. My interests include:

  • 🔧 LLMs for Code: Investigating how to advance the capabilities of large language models in code generation, automated program repair, and software security. I’m particularly interested in methods that improve code reliability and maintainability, as well as techniques for modeling and mitigating adversarial behaviors in LLM-powered code systems.

  • 🧠 LLM Reasoning: Aiming to enhance the reasoning efficiency and fidelity of LLMs by developing approaches that reduce computational overhead while preserving or improving logical consistency and correctness. My focus lies in enabling LLMs to reason more accurately, coherently, and efficiently under constrained inference settings.

I’m always excited about discussions and collaborations! Feel free to reach out via email if you’re interested in my work or just want to chat about research ideas 💬.

Education

  • 🎓 B.Sc. in Computer Science, 2020 – 2024, Nankai University, Tianjin, China
  • 🎓 M.Sc. in Artificial Intelligence, 2024 – Present, University of Southern California, Los Angeles, CA, United States

Selected Publications

  1. IEEE S&P
    Your Compiler is Backdooring Your Model: Understanding and Exploiting Compilation Inconsistency Vulnerabilities in Deep Learning Compilers
    Simin Chen, Jinjun Peng, Yixin He, and 2 more authors
    2025
  2. emnlp2025
    Recent Advances in Large Langauge Model Benchmarks against Data Contamination: From Static to Dynamic Evaluation
    Simin Chen, Yiming Chen, Zexin Li, and 8 more authors
    2025
  3. emnlp2025
    Diagnosing Memorization in Chain-of-Thought Reasoning, One Token at a Time
    Huihan Li, You Chen, Siyuan Wang, and 4 more authors
    2025