Jiajun Xu
Graphics and Geometric Computing Group, Room 3 - 523, Information Technology Building (FIT), Tsinghua University
Research Interests
- My research interests lay on machine learning, computer vision and multi-modal. Specifically, I'm passionate about exploring visual model understanding and visual generation. My specific research interests are:
- Reinforcement learning for video generation models, manipulating diffusion model noise to get faster and stronger generators.
- Exploring the latent space of transformers and fine-tuning attention to reduce multi-modal hallucinations.
Education
Tsinghua University, Department of Computer Science and Technology
Beijing, China
September 2022 – June 2026
- GPA：3.69/4.0
- Major Courses: Computer Language and Programming, Introduction to Computer Systems, Data Structures, Discrete Mathematics, Object-Oriented Programming, Software Engineering, Computer Graphics, Principles of Compilation, Principles of Signal Processing, Artificial Neural Network, Reinforcement Learning
- Supervisor: Shi-Min Hu
Publication
R-Bench: Graduate-level Multi-disciplinary Benchmarks for LLM & MLLM Complex Reasoning Evaluation
Meng-Hao Guo, Jiajun Xu, Yi Zhang, Jiaxi Song, Haoyang Peng, Yi-Xuan Deng, Xinzhi Dong, Kiyohiro Nakayama, Zhengyang Geng, Chen Wang, Bolin Ni, Guo-Wei Yang, Yongming Rao, Houwen Peng, Han Hu, Gordon Wetzstein, Shi-min Hu
arXiv (ICML 2025)
VisionReward: Fine-Grained Multi-Dimensional Human Preference Learning for Image and Video Generation
Jiazheng Xu*, Yu Huang*, Jiale Cheng, Yuanming Yang, Jiajun Xu, Yuan Wang, Wenbo Duan, Shen Yang, Qunlin Jin, Shurun Li, Jiayan Teng, Zhuoyi Yang, Wendi Zheng, Xiao Liu, Ming Ding, Xiaohan Zhang, Xiaotao Gu, Shiyu Huang, Minlie Huang, Jie Tang, Yuxiao Dong
arXiv (ICCV 2025 under review)
Skills
Artificial Intelligence: PyTorch, Transformer Frameworks, Diffusion Model, build model frameworks and train models.
Programming Languages: C, C++, Python, Java, Rust.
Research Experience
Graphics and Geometric Computing Group, Tsinghua University
Beijing, China
October 2024 – Present
- Conducted research in the fields of computer graphics, computer vision, and multimodal large models under the guidance of Professor Shi-Min Hu.
- Joined the RBench project under the guidance of Meng-Hao Guo, a PhD candidate from GGCG Lab.
- Leading the Reasoning Attention project conworking with Meng-Hao Guo.
RBench Project
Beijing, China
November 2024 – February 2025
- Created a benchmark that can test the reasoning capabilities of models, covering multimodal, multilingual, and interdisciplinary aspects.
- Built a model evaluation framework suitable for this project using VLMEvalKit.
- The research findings have been accepted by ICML 2025.
Reasoning Attention Project
Beijing, China
March 2025 – Present
- Lead a project coworking with Meng-Hao Guo, aiming to reducing the multi-modal hallucinations caused by the low attention of the visual tokens.
- The research findings are preparing for AAAI 2026.
AI Research Intern, Beijing Zhipu AI Technology Co., Ltd.
Beijing, China
March 2024 – April 2025
- Actively involved in a video generation model project as a research intern at Zhipu AI.
- Experienced in distributed training with hundreds of GPUs and proficient in training with CogVLM.
Knowledge Engineering Group, Tsinghua University
Beijing, China
November 2023 – December 2024
- Participated in multimodal group at KEG Lab, focusing on textual-visual graph research under the guidance of Associate Professor Yuxiao Dong.
- Joined the VisionReward project under the guidance of Jiazheng Xu, a PhD candidate from KEG Lab, aiming to develop a reward model for vision generative models.
VisionReward Project
Beijing, China
June 2024 – December 2024
- Developed a reward model for evaluating image and video quality precisely and explainably in different dimensions.
- Conducted experiments, took on engineering tasks within the project, and found an appropriate way to train a reward model.
- The final outcome of VisionReward is under review with ICCV 2025.
Tsinghua Information Retriever Group, Tsinghua University
Beijing, China
May 2024 – October 2024
- Focused on research about Retrieval-augmented Generation under the guidance of Assistant Professor Qingyao Ai.
- Constructed a RAG system with LangChain framework.
Contest Experience
Tsinghua University Intelligent Agent Competition, Programmer
Beijing, China
May 2023
- Self-taught Python, Git, computer networks, etc., and collaborated with a team to develop an intelligent agent for the game "Minecraft" that can automatically perform player operations.
2024 COMAP Mathematical Contest in Modeling, Team Leader
Beijing, China
February 2024
- Planned and coordinated the team's work, led the team to study the necessary knowledge, and received an Honorable Mention Award.
- Used multi-level analysis, K-means algorithm, and TOPSIS comprehensive evaluation method to model the risk level of a region, quantifying a risk index directly usable by insurance companies.
Extracurricular Activities
Class Monitor of Class 22 in the Department of Computer Science and Technology, Tsinghua University
Beijing, China
September 2023 – September 2024
- Managed class affairs, assisted class teachers and counselors in class management, and organized various class and group activities.
- During tenure as class monitor, the class was awarded the title of Best Youth League Branch, the highest honor for a class in Tsinghua University.
Vice Minister of Department of Internal Contacts, Tsinghua University Science Association
Beijing, China
September 2023 - Present
- Coordinated and organized various scientific research competitions within the university.
- Established connections between different departments and the university science association.