Xirui Wang

Xirui Wang

Undergraduate Researcher, Huazhong University of Science and Technology

Email | GitHub | CV

About

I am an undergraduate student in Robotics Engineering at Huazhong University of Science and Technology (HUST), expected to graduate in 2028.

I am interested in how intelligent systems acquire structured understanding of the world—moving beyond surface-level pattern recognition toward models that reason, adapt, and interact in a principled way. Currently, I am exploring structural anomaly detection, controllable image generation, and the application of reinforcement learning in diffusion models.

I am actively seeking research opportunities for Summer 2026 and would be grateful for any consideration or advice.

News

Research Interests

Structural Representations Controllable Generation Reinforcement Learning Diffusion Models

Education

Huazhong University of Science and Technology 2024 – 2028 (Expected)
B.Eng. in Robotics Engineering
Relevant coursework: Machine Learning, Data Science, Probability and Statistics

Research Experience

Independent Research 2025 – Present
School of Mechanical Science and Engineering, HUST
  • Authored a comprehensive survey on UAD, covering data generation, backbones, and training strategies, with discussion of current limitations and emerging research trends.
  • Independently proposed a training-free instance-level anomaly detection framework by integrating multi-cue information with interpretable outputs; achieved strong performance among unsupervised methods on MVTec LOCO.
Research Intern 2025 – Present
School of Computer Science and Technology, HUST
  • Worked on I2E for interactive multi-layer image editing; built evaluation dataset pipeline and contributed to the "Decompose-then-Action" pipeline design.
  • Independently exploring efficient RL-based post-training for flow models by combining flow dynamics with group-based advantage optimization (e.g., GRPO).

Publications & Preprints

Unsupervised Anomaly Detection in Industry: Progress, Trends, and Future Avenues
Xirui Wang, et al.
Under review
I2E: From Image Pixels to Actionable Interactive Environments for Text-Guided Image Editing
Jinghan Yu, Junhao Xiao, Chenyu Zhu, Jiaming Li, Jia Li, HanMing Deng, Xirui Wang, et al.
arXiv:2601.03741, 2026.

Selected Projects

LazyDDL: AI-powered Deadline Assistant
A prototype app with OCR/ASR deadline extraction, auto-generated tasks and reminders, and lightweight analytics. Designed for users overwhelmed by dense workloads and hard-to-track deadlines.
Qiming Mirror: Multisensory AI Assistant for the Visually Impaired
A demo system converting visual inputs into multisensory descriptions using OCR, object recognition, and a lightweight RAG module, enabling safer daily decisions for visually impaired users.

Technical Skills

Technical: Python, C++, PyTorch, Linux, Git, Docker

Research: Computer Vision, Diffusion Models, Reinforcement Learning