About Me

I build learning systems that move between geometry, scientific discovery, and real-world decisions.

I am an incoming master's student at the School of Software, Tsinghua University, advised by Prof. Mingsheng Long. I received my B.E. from the same school, with a GPA of 3.9/4.0 and a class rank in the top 6%.

My research explores machine learning systems that connect representation learning, scientific computing, and intelligent agents. I care about both the ideas behind learning algorithms and the engineering needed to make them reliable and useful.

I welcome research discussions and collaborations. Feel free to reach out by email.

Research Interests

My current interests include:

• Scientific Machine Learning: neural PDE solvers, geometric representations, and efficient learning systems.
• Video Generation: generative video models, visual dynamics, and world modeling.
• Physical AI: learning systems that perceive, reason about, and interact with the physical world.

News

• [Jun. 2026] Brep2Shape is accepted by ICML 2026.
• [Dec. 2025] FlashBias is accepted by NeurIPS 2025.
• [Sep. 2025] Completed my internship at JingDong, working on the open-source agent framework OxyGent.
• [Feb. 2025] Joined THUML's Neural Solver Library project as a major contributor.

Education

Publications

1. ICML 2026

   Brep2Shape: Boundary and Shape Representation Alignment via Self-Supervised Transformers

   Yuanxu Sun*, Yuezhou Ma*, Haixu Wu#, Guanyang Zeng, Muye Chen, Jianmin Wang, Mingsheng Long#

   International Conference on Machine Learning, 2026.

   arXiv PDF
2. Under Review

   Feasible Policy Optimization for Safe Reinforcement Learning

   Yujie Yang*, Yuanxu Sun*, Wenyu Li, Beiyan Jiang, Tao Zhang, Shengbo Eben Li#

   Under review.

   Public link pending
3. NeurIPS 2025

   FlashBias: Fast Computation of Attention with Bias

   Haixu Wu, Minghao Guo, Yuezhou Ma, Yuanxu Sun, Jianmin Wang, Wojciech Matusik, Mingsheng Long#

   Advances in Neural Information Processing Systems, 2025.

   arXiv PDF Code

* Equal contribution   # Corresponding author

Selected Projects

• Neural Solver Library ★ 322

  An open-source library for neural PDE solvers. I contributed to the infrastructure, model reproduction, systematic experiments, and performance analysis. I optimized 22 underperforming experiments and resolved 19 errors.

  Code

• OxyGent ★ 2k

  An open-source multi-agent system framework for complex industrial workflows. As a core developer, I worked on agent design, flow architecture, project management, and the Negotiation Specialist pipeline.

  Code Paper Docs

Experience

Research Experience

Industry Experience

Honors & Awards

• [2025] Comprehensive Excellence Scholarship
• [2025] 3rd Prize (North China Region), National Software Creativity Competition
• [2024] Comprehensive Excellence Scholarship
• [2024] 1st Prize, Tsinghua University Software Innovation Competition
• [2024] Outstanding SAST Member, School of Software
• [2023] Academic Excellence Scholarship

Skills & Service

• Programming: Python, C++, MATLAB
• Frameworks & Tools: PyTorch, Git, LaTeX
• Languages: Chinese, English
• Community: Class President; Vice President, SAST Competition Department; Tsinghua Red Cross Society.