Zongxian Yang
I am a Research Assistant at the Healthcare Analytics & Neural Intelligence (HANI) Lab, City University of Hong Kong (Dongguan). I received my B.Eng. in Electronics and Information Engineering from Huazhong University of Science and Technology in 2023, and I will join the Department of Computer Science at City University of Hong Kong as a PhD student in June 2026.
My research focuses on LLM reasoning, trustworthy medical AI, medical agents, and multimodal reasoning. I am particularly interested in building systems whose reasoning process is not only strong in performance, but also verifiable, correctable, and auditable.
Research Interests
- LLM reasoning beyond surface chain-of-thought performance
- Verifiable, self-correctable, and auditable reasoning processes
- Medical LLMs and trustworthy medical AI
- Clinical workflow abstraction with agent or multi-agent systems
- Multimodal reasoning and medical VLMs
- Synthetic data, reflection, and self-evolving reasoning systems
Selected Research
Med-REFL: Medical Reasoning Enhancement via Self-Corrected Fine-grained Reflection
This work studies how fine-grained reflection and self-correction can improve the reliability of medical reasoning, with a Tree-of-Thought style structure for building preference data and training signals (first author).
QM-ToT: A Medical Tree of Thoughts Reasoning Framework for Quantized Model
This project explores how to preserve or improve medical reasoning performance under quantization and resource constraints (first author).
Better Eyes, Better Thoughts: Why Vision Chain-of-Thought Fails in Medicine
This project analyzes why vision chain-of-thought often fails in medical settings and studies the limits of multimodal medical reasoning (co-first author).
Hyperbolic Relational Prompts for Intersectional Fairness in Medical VLMs
This project studies fairness in medical VLMs through relational prompting and representation design (co-author).
News and Updates
- June 2026: Joining the Department of Computer Science at City University of Hong Kong as a PhD student.
- April 2026: Hyperbolic Relational Prompts for Intersectional Fairness in Medical VLMs was accepted by CVPR as an Accept (Highlight) paper.
- March 2026: Released Better Eyes, Better Thoughts on arXiv and submitted the work to MICCAI.
- June 2025: Released Med-REFL on arXiv.
- April 2025: Released QM-ToT on arXiv.