CV

Education

B.S.E. in Computer Science
- University of Michigan, Ann Arbor
- Aug 2025 - Present
- GPA: 3.88/4.00
- Relevant Coursework: Machine Learning, Intro to Algorithmic Robotics (A), Operating Systems, Applied Parallel Programming with GPUs
B.Eng. in Mechanical Engineering
- Shanghai Jiao Tong University
- Aug 2023 - Present

Robot Learning (VLA, VAM), Agentic Systems, and Efficient Machine Learning.

Research Assistant (Nov 2025 – Present)
- Efficient Inference for Embodied Foundation Models
- Advisor: Dr. Jiachen Liu (University of Michigan)
- Profiled COSMOS-Policy across denoising steps; identified that block-level residual skipping is infeasible while cross-attention outputs remain highly stable (cosine > 0.999).
- Validated cross-attention KV caching on 24 RoboCasa tasks: achieved identical task success rate of 68.06% to the baseline across 3,600 rollout trials with consistent denoising speedup across all 24 tasks.
- Designing task-aware token compression that exploits the model’s latent-frame slot structure to selectively prune image tokens via self-attention reduction, targeting further denoising acceleration on top of the validated caching strategy.
Research Assistant (Sep 2024 - Aug 2025)
- Gravitational Effect on Swarming Behavior of Microorganisms
- Advisor: Prof. Zijie Qu (Shanghai Jiao Tong University)
- Trained a U-Net model for colony boundary segmentation; diagnosed systematic failure modes, then adopted a SAM-based pipeline that achieved reliable automated detection, replacing manual annotation.

Web Agent Version-Robust Benchmark (Jan 2026 - Present)
- EECS 545 (Machine Learning) Course Project
- Advisor: Prof. Honglak Lee
- Built a reproducible testing pipeline using Docker to deploy baseline agents like QWen3-VL-30B across historical snapshots of open-source websites like SimpleWiki.
- Formulated a taxonomy for web version variations and designed controlled experiments to isolate version-induced failure modes in LLM-based web agents.
- Implementing a Knowledge Graph-based method to automate benchmark task generation across website versions.
Probabilistic Motion Planning for Redundant Robots (Nov 2025 - Dec 2025)
- EECS 465 (Intro to Algorithmic Robotics) Course Project
- Advisor: Prof. Dmitry Berenson
- Reproduced and benchmarked sampling-based motion planning algorithms (RRT-Connect, PRM) for a 7-DOF Franka Panda robot in simulation; implemented a hybrid sampling strategy that reduced trajectory generation latency by ~75%.

Languages: Mandarin (native), English (fluent, TOEFL 100)
Programming Languages: Python, C/C++, MATLAB, Java, LaTeX, Typst
Technical Skills: Hugging Face, PyTorch, Git, Claude Code, Linux, Docker, ssh, SolidWorks