Haohan Yuan
I study how structural priors can make LLM, multimodal, and agentic systems more reliable, explainable, and controllable.
I am a Computer Science Ph.D. student at the University of Hawaii at Manoa, advised by Prof. Haopeng Zhang. In August 2026, I will transfer to the University of North Carolina at Charlotte with my advisor, with my Ph.D. coursework and qualifying exam transferred. My research spans LLM reasoning, long-context understanding, model attribution, multimodal systems, and embodied agents.
Academic path: B.Sc. in Mathematics and Applied Mathematics, Hefei University of Technology (2018–2022) · M.Eng. in Computer Science, Nanyang Technological University (2022–2024), advised by Prof. Siu Cheung Hui · Ph.D. study in Computer Science, University of Hawaii at Manoa (August 2024–July 2026) · Incoming Ph.D. student, UNC Charlotte (August 2026).
🤖 Research Highlights
Extending structured reasoning and reliability from language models to multimodal and embodied systems.
StrucVLA
Structural Priors for Explainable Vision-Language-Action Policies
Exposing high-level VLA understanding as explicit scene, task, interaction, and progress states that can be supervised, diagnosed, and controlled.
SumFingerprint
Reliability and Fragility of LLM Summary Fingerprints
Investigating whether model-specific fingerprints in generated summaries remain trustworthy under rewriting and targeted transformation attacks.
From structured language reasoning to reliable AI systems
My published work studies structured reasoning, faithfulness, and robustness in summarization. I am now carrying those ideas into agentic AI and vision-language-action policies.
View research figure
📚 Publications
Position: Generative Engine Optimization Creates Underexamined Risks
Identifies governance risks involving concentrated influence, disclosure, and evaluation blind spots.
Paper✨ Recent News
Our position paper on the risks and governance of Generative Engine Optimization was accepted.
I will transfer to the University of North Carolina at Charlotte with my advisor in August 2026, with my Ph.D. coursework and qualifying exam transferred.
I received an offer for the 2026 fellowship from the Hawaii Data Science Institute.
Our work on LLM reasoning for abstractive summarization was accepted.
Our StrucSum paper on graph-structured long-document reasoning was accepted.
I received the 2025 fellowship from the Hawaii Data Science Institute.
Our work was featured in University of Hawaii News.
Our DomainSum paper on fine-grained domain shift in summarization was accepted.
I joined the University of Hawaii at Manoa as a Computer Science Ph.D. student.
Interested in reliable LLMs or structured multimodal systems?
I am always happy to discuss research ideas, collaborations, and internship opportunities.