Hello 👋, I’m Hongrui Jia (贾泓睿). I’m a second-year Master’s student at Peking University, advised by Prof. Shikun Zhang and Prof. Wei Ye. My primary research interests lie in Multimodal Large Language Models (MLLMs) and MLLM-based agents. I have published several papers at top-tier international AI conferences and journals with total google scholar citations 179. If you’re interested in collaboration, feel free to reach out to me at jiahongrui@stu.pku.edu.cn.
📚 Educations
- 2024.09 - 2027.06, MS, Software Engineering, Peking University.
- 2020.09 - 2024.06, BS, Software Engineering, South China University of Technology.
🔬 Research Interests
My research focuses on advancing trustworthy and capable Multimodal Large Language Models (MLLMs) through rigorous evaluation, targeted alignment, and diagnostic-driven training paradigms, with a commitment to building reliable, open-source multimodal intelligence systems.
This work follows a progressive research pipeline: systematically identifying critical failure modes in MLLM reasoning, comprehensively exploring the boundaries of multimodal reasoning capabilities.
Systematic Evaluation of Critical Weaknesses in Reasoning. The reliability of MLLM reasoning is constrained by fundamental bottlenecks: cognitive drift induced by hallucinations and structural deficiencies in tool-calling capabilities. Hallucinations not only introduce factual errors but also trigger cascading deviations throughout reasoning chains; insufficient tool invocation abilities fundamentally limit models’ capacity for deep interaction with external knowledge sources and execution environments. How can we construct fine-grained evaluation frameworks to quantify these latent deficiencies? How might we establish comprehensive benchmarks covering visual consistency, tool coordination, and decision reliability?
Related Works: Hal-Eval (ACM MM 2024), OSWorld-MCP (ICLR 2026).
Precision Enhancement and Dynamic Evolution of Reasoning Capabilities. How can we compel models to genuinely “see” rather than merely “read” visual evidence during complex reasoning? How do we transcend the limitations of single-image understanding to achieve information integration and relational reasoning across multiple images? Within extended reasoning chains, what mechanisms can anchor visual grounding to prevent logical drift? Most critically, how can we translate diagnostic insights into real-time adjustments of data generation and training strategies, enabling continuous, targeted capability evolution?
Related Works: SymDPO (CVPR 2025), MaVEn (NeurIPS 2024), Decoupled Reasoning and Perception (Preprint), DPE (Preprint).
📝 Publications
OSWORLD-MCP: BENCHMARKING MCP TOOL INVOCATION IN COMPUTER-USE AGENTS
Hongrui Jia, Jitong Liao, Xi Zhang, Haiyang Xu, Tianbao Xie, Chaoya Jiang, Ming Yan, Si Liu, Wei Ye, Fei Huang
🎖 Honors and Awards
- 2025.10 National Scholarship.
- 2021.10 National Scholarship.