Yu Xinqiang

Yu Xinqiang

Undergraduate in Cyberspace Security

I am Xinqiang Yu (喻新强), an undergraduate student in Cyberspace Security at Tianjin University (2024 – 2028). My research interests focus on AI for AI, LLM Safety, and Multi-Agent Systems. I enjoy building systems that connect rigorous reasoning with practical engineering.

My research interests include:

  • AI for AI — automating AI research with AI. My work includes AutoSFT, a multi-agent harness that automates SFT data research by iteratively rewriting the training data recipe while the training stack stays frozen, and equiNAS, a typed DSL for certified evolution of equivariant architectures, where every LLM-proposed mutation is machine-checked for type safety, equivariance, and dimension consistency before any GPU time is spent.
  • LLM Safety — post-training safety alignment of large language models: safety-aware preference optimization, personalized safety adaptation, and multi-turn jailbreak robustness, explored through a research internship on personalized LLM safety at China Mobile.
  • Multi-Agent Systems — multi-agent collaboration for security and competition. Previously: an adversarial multi-agent scoring system (Chinese Collegiate Computing Competition, 4C) and multi-agent RL (NeurIPS Lux AI Season 3, Kaggle bronze). Currently: building an AutoHunter-style multi-agent system for automated SRC vulnerability discovery, with specialized agents for target collection, tool-driven reconnaissance, exploitability review, and mass-exploitation checking — humans only as final reviewers.

📰 News

📄 Paper News

  • 2026.07 - 2026.09 Visiting student at Zhongguancun Academy, Beijing.
  • 2026.04 🎉 Our paper “Beyond Retrieval: Bi-Temporal State Arbitration for Longitudinal Healthcare Agents” was published at the KnowFM Workshop @ ACL 2026. [Paper]
  • 2026.03 Completed a research internship on personalized LLM safety and trustworthy alignment at China Mobile.
  • 2025.10 Joined TankLab at Tianjin University, working on inference performance analysis of text-to-video generation models and multi-model hybrid deployment for inference and training.

🏆 Competition News

  • 2026.8 🥉 National Third Prize in the 9th China Collegiate Intelligent Robot Creative Competition.
  • 2026.8 🥉 National Third Prize in the 19th Chinese Collegiate Computing Competition (4C).
  • 2026.7 🥇 First Prize (finals) in the 23rd Information Security and Countermeasures Contest (ISCC).
  • 2026.5 🥇 First Prize in the Huawei Software Elite Challenge (Beijing-Tianjin-Northeast China Region), advancing to the global finals.
  • 2026.5 🏅 Meritorious Winner in the Mathematical Contest in Modeling (MCM).
  • 2026.4 🥈 National Second Prize in the Practice Competition (Ascend AI Track) of the 10th Huawei ICT Competition.
  • 2025.3 🥉 Bronze Medal in the NeurIPS 2024 - Lux AI Season 3 competition on Kaggle, reaching the global top 15.
  • 2024.12 🥇 First Prize in the 16th Chinese Mathematics Competitions (CMC), ranked 7th in Tianjin.

🎓 Education

  • 2024.09 - 2028.06 (expected), Tianjin University, B.Eng. in Cyberspace Security, Tianjin, China
    • Highest single-semester weighted score: 95.49 / 100; academic-year ranking: 2/102 by weighted average, 1/102 by comprehensive evaluation
    • Core courses: Natural Language Processing, Algorithm Design and Analysis, Database Systems, Computer Networks, Computer Organization, Operating Systems

📝 Publications

  • Jianing Zhao, Xiaoquan Zhi, Xinqiang Yu. Beyond Retrieval: Bi-Temporal State Arbitration for Longitudinal Healthcare Agents. KnowFM Workshop @ ACL 2026.
    [Paper] [PDF]

💻 Experience

  • 2026.07 - 2026.09, Visiting Student, Zhongguancun Academy, Beijing
  • 2025.10 -, Research on inference performance analysis of text-to-video models and multi-model hybrid deployment for inference and training, TankLab, Tianjin University
  • 2026, Research Intern (online), China Mobile — personalized LLM safety and trustworthy alignment, concluded 2026.03