Qunzhong Wang

Qunzhong Wang (王群中)

Hi, I'm Qunzhong Wang. I am currently an undergraduate at the Chinese University of Hong Kong, majoring in Mathematics and Information Engineering.

Prior to my undergraduate studies, I received the gold medal in Chinese Mathematical Olympiad (aka CMO).

My current research interests includes:

Principles of AI Systems backed by Math

Understanding the mathematical principles behind model representation capacity, training dynamics, and generalization.
Leveraging these principles to design better and more scalable architectures, optimizers, training/fine-tuning methods, and regularization techniques.

,

Reinforcement Learning on Large Models

Aligning Large Language Models (LLMs), Vision-Language Models (VLMs), and their derivative Agents with specific human preferences and demands, with techniques like Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning with Verifiable Reward (RLVR).
Exploring robust fine-tuning "recipes" within the RL framework to ensure that pre-trained capabilities are preserved while desired, human-aligned skills are effectively amplified.

I feel incredibly fortunate to have been supervised by distinguished scholars: Hong Cheng; Xiangyu Yue; Sotirios Sabanis; Zhuang Liu; Weiyang Liu.

You can view my full resume here.

Email / Scholar / Github

Publications

Does Graph Prompt Work? A Data Operation Perspective with Theoretical Analysis
Qunzhong Wang, Xiangguo Sun, Hong Cheng
International Conference on Machine Learning (ICML), 2025

arXiv / Paper / Code

This paper introduces a theoretical framework that rigorously analyzes graph prompting from a data operation perspective. The error lower bound, upper bound, and data distribution have been thoroughly studied.

VideoSearch Reasoner: Boosting Multimodal Reward Models through Thinking-with-Image Reasoning
Qunzhong Wang, Jie Liu, Jiajun Liang, Yilei Jiang, Yuanxing Zhang, Yaozhi Zheng, Xintao Wang, Pengfei Wan, Xiangyu Yue, Jiaheng Liu
Preprint, Under Review at International Conference on Learning Representations (ICLR) 2026

arXiv / Paper / Code

This work breaks the inherent limitations of VLMs in handling the number of video frames by adding tool invocation capabilities to the multimodal Reward Model, greatly enhancing factual accuracy and significantly surpassing the baseline.

ScreenCoder: Advancing Visual-to-Code Generation for Front-End Automation via Modular Multimodal Agents
Yilei Jiang, Yaozhi Zheng, Yuxuan Wan, Qunzhong Wang, Jiaming Han, Michael R. Lyu, Xiangyu Yue
Preprint, Under Review at International Conference on Learning Representations (ICLR) 2026

arXiv / Paper / Code

This work achieved end-to-end generation of front-end to front-end code, implementing a revolutionary breakthrough.

Selected Awards

	Talent Development Scholarship, HKGOV (2025)
	Professor Charles K. Kao Research Exchange Scholarships, Chinese University of HK (2025)
	Dean's List, Chinese University of HK (2024, 2025)
	ELITE Stream Scholarship, Chinese University of HK (2024, 2025)
	Mathematical Modeling Contest, Meritorious Winner (2024)
	Soong Ching Ling Scholarship (2023)
	Prof Omar Wing Memorial Scholarship (2023)
	Simon Marais Mathematics Competition, 11th in East Division (2023)
	Admission Scholarship, Chinese University of HK (2023)
	China Physics Olympiad (Provincial), First Prize (2022)
	China Chemistry Olympiad (Provincial), First Prize (2022)
	China Mathematics Olympiad, Gold Medal (2022)

Experience

Services

	International Conference on Learning Representations (ICLR) Reviewer, 2026
	IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Reviewer, 2026

Internship

	2024.04—2024.09: Database Research Group, CUHK, advised by Prof. Hong Cheng
	2024.12—2025.09: Kling AI Technology Department, Kuaishou
	2025.09—Present: Zhuang's Lab, Princeton, advised by Prof. Zhuang Liu