Beijie Liu

About Me

I am a Ph.D. student in Computer Science at the University of Southern California in Sept Lab. My work focuses on building secure and reliable AI systems, with an emphasis on LLM verification, GPU/CPU co-execution, and system-level reliability. I enjoy working across the boundary of machine learning and systems, especially when it involves understanding how models run on real hardware and how they fail in subtle, hard-to-detect ways.

Previously, I worked on training-time bug detection at the University of Michigan in OrderLab and on speech-related ML models in CHAI Lab.

Research Interests

• LLM Security & Verification: ensuring trustworthy inference, detecting cheating/incorrect execution, and building lightweight verification layers
• Systems for ML: GPU/TEE co-execution, confidential inference, and performance analysis
• Reliable Machine Learning: silent error detection, robust training, and model debugging
• General ML: representation learning, multimodal workloads, and model behavior analysis

News

• [Dec. 2025] Our paper WAVE: Leveraging Architecture Observation for Privacy-Preserving Model Oversight is accepted to ASPLOS 2026.
• [Sept. 2025] I joined the Sept Lab at USC as a Ph.D. student.
• [Mar. 2025] Our work Training with Confidence is accepted to OSDI 2025.

Publications

Powered by Jekyll and Minimal Light theme.