I hold full-stack AI expertise spanning AI Infrastructure, Physical Intelligence and MLLMs, and my research explores the cutting-edge frontiers and practical societal value of artificial intelligence, with a core focus on Trustworthy AI, AI for Healthcare, and safety across LLMs, VLMs, WAMs and autonomous agents.

🔥 News

• 2026 Currently working on trustworthy AI, self-evolving agents, multimodal reinforcement learning, and embodied intelligence grounded in physical interaction.
• 2026 Actively participating in scientific research and internship opportunities, with sustained interest in reliable, deployable, and safety-critical AI systems.
• 2025-2026 Continuing first-author research on deployment reliability, compiler-triggered backdoor risks, and multimodal medical AI.

📝 Publications

ASE 2026

PrecisionDiff: Automated Detection of Precision-Induced Disagreements in Large Language Models

Yifei Wang, collaborators with Shanghai Jiao Tong University, Nanyang Technological University, and Beihang University

Paper

• First author, under review at ASE 2026.
• An automated testing pipeline for detecting cross-precision disagreements in large language models.
• Focuses on deployment-aware safety evaluation under practical inference settings.

AAAI 2027

Reliability Testing of Medical Model Performance under Distributed Deployment

Yifei Wang, with advisors Tianlin Li, Xiaohan Zhang, and Xiaoyu Zhang

• First author, in preparation for AAAI 2027.
• Studies how distributed frameworks such as vLLM and DeepSpeed induce backend-dependent behavior divergence.
• Focuses on safety-critical medical deployment risks caused by operator fusion, parallel reduction, and precision truncation.

AAAI 2027

Trusted Weights, Treacherous Backends: Compiler-Triggered Backdoor Attacks

Yifei Wang, with advisors Tianlin Li, Xiaohan Zhang, and Xiaoyu Zhang

• First author, in preparation for AAAI 2027.
• Investigates compiler-triggered backdoor behavior that does not reside in model weights.
• Reveals a new attack vector that can evade static weight-based safety audits.

🎖 Honors and Awards

• 2025 National Scholarship, ranked 1/136.
• 2025 Outstanding Students, Tongji University.
• 2024 First-Class Scholarship for Outstanding Undergraduate Students, Tongji University.
• 2023 Freshmen Scholarship.
• 2023-2026 Received more than 10 awards at national, provincial, and university levels.

📖 Educations

• Sep 2023 - Present, Tongji University, Computer Science and Technology (Elite Class 2023), Shanghai, China.
• Current GPA: 4.92/5, Rank: 3/125.

🏆 Competitions

• 2025, China International College Students’ Innovation Competition
  • Silver Award, Shanghai Regional (Founder and Team Leader)
  • Gold Award, Shanghai Regional (Team Member)
• 2024, RoboCup China National Competition
  • National First Prize
  • Project focus: lunar robot navigation and robotic arm grasping
• 2026, National AI Application Scenario Innovation Challenge
  • National Second Prize, Intelligent Medical Special Track
  • Team Leader
• 2025, National Undergraduate Embedded Chip and System Design Contest
  • Eastern China Regional Second Prize
  • Lead algorithm developer for UAV anomaly detection

🌟 Comprehensive Quality

• 2025.07 Probationary Member of the Communist Party of China
  • Participated in volunteer work, academic lectures, and student activities promoting scientific spirit.
• 2025.09 Vice Director, Cross-campus Office, Guohao Academy, Tongji University
  • Coordinated cross-campus academic events and student services.
• 2025.07 Summer Research Exchange: ETH Zurich and EPFL
  • Engaged in academic exchange related to advanced AI systems.
• Ongoing Sport and Music
  • Guitarist; gold medalist in university athletics; member of Tongji Football League and FA Cup.
• Ongoing Academic Conferences and Independent Learning
  • Self-driven study in AI safety and computer vision, with active participation in seminars and workshops.
• Ongoing Social Media Creator Xiaohongshu (Red)
  • Share research notes and technical insights in AI.
  • Treat documentation as an essential part of deep understanding and academic outreach.