CV

Education

Wuhan University, B.S. in Electronic Information Engineering, 2024

  • GPA: 3.60/4.0 (87.2)
  • Third Class Scholarship of School of Electronic Information (2021)

Research Experience

Rutgers University, New Brunswick<br /> Visiting Student, advised by Prof. Hao Wang | Mar 2024 - Mar 2025<br /> Topics: Recommender Systems, Multimodal Learning

University of Illinois Urbana-Champaign<br /> Research Intern, advised by Prof. Tong Zhang | Mar 2025 - Jan 2026<br /> Topics: Multimodal Learning

University of California, San Diego<br /> Research Intern, advised by Prof. Biwei Huang | Sep 2025 - Jan 2026<br /> Topics: World Models, Embodied AI

Nanyang Technological University<br /> Research Intern, advised by Prof. Yang Liu | Apr 2026 - Present<br /> Topics: Embodied AI, Physical Reasoning

Research Projects

Task-Adaptive Attention and Memory for Embodied VLM Reasoning<br /> Advisor: Yang Liu | Apr 2026 - Present

  • Designed a task-adaptive visual attention mechanism that selectively enhances attention over question-relevant objects and image regions for robot-oriented VLM tasks.
  • Developed a hybrid short-term and long-term memory framework that retrieves task-relevant context and dynamically updates long-term memory for context-aware visual reasoning.

Latent-Dynamics World Model with VLM-Guided Data Filtering<br /> Advisor: Biwei Huang | Oct 2025 - Jan 2026

  • Modeled predicted latent representations as states and their temporal differences as actions, adding a latent-space action loss to improve trajectory consistency.
  • Used Qwen3VL-32B as a vision-language judge to filter high-quality synthetic samples and form a data flywheel.

Physics Diagram Synthesis<br /> Advisor: Tong Zhang | Oct 2025 - Jan 2026

  • Proposed a method for generating synthetic physics problem solutions using reinforcement learning with augmented feedback.
  • Leveraged Qwen2.5VL-7B to iteratively improve reasoning and visualization generation through automated evaluation and data selection.

Interpretability of MLLMs<br /> Advisor: Prof. Hao Wang | Mar 2024 - Sep 2024

  • Identified strong attention correlations between visual inputs and token-level outputs in LLaVA.
  • Proposed an adaptive pruning technique for hierarchical attention layers, improving accuracy and inference efficiency.

Skills

  • Programming: Python, C++, C, Lean, PyTorch
  • Tools: ROS, Docker, Gazebo, IsaacSim

Selected Publications

See the Publications page for a complete list of research publications.