😊 About me
Contact: vanzl3386 [at] gmail.com (preferred), vanzl [at] u.nus.edu, 121090525 [at] link.cuhk.edu.cn
Hi! I am Zhenglin Wan (万政霖), a CS Ph.D. student at HPC-AI Lab at National University of Singapore (NUS), advised by Prof. Yang You. Previously, I worked at Nanyang Technological University (NTU) with Prof. Bo An, Centre for Frontier AI Research (CFAR), IHPC, A*STAR with Prof. Ivor Tsang , Hong Kong Generative AI Research & Development Center (HKGAI) with the team led by HKUST Prof. Yike Guo. I received my B.Sc from Chinese University of Hong Kong (CUHK) on 2025 Fall, during which I served as tech co-founder of AI start-up Metasequoia Intelligence based in Shenzhen, China.
I am currently interested in:
- Post-training of Visual Generation Models, specifically, understanding the dynamics of diffusion-transformer and the synergy and conflict of standard RL paradigm.
- Interactive Video Generation and the extension to world models.
- Agentic Systems. However, I am focusing more on the “agentic sense”, rather than viewing “agentic AI” as a research domain. As I believe that large video models or world models could also benefit from agentic design in the future, as they are the same with LLMs from the foundamental perspective.
Always enjoy collaborating with talents, plz reach out if you are interested!
🔥 News
- 2026.04 🎉🎉 Invited talk by Meta MRS (Stable online RL for alignment of diffusion-based foundation models)!
- 2026.03 🎉🎉 Two papers accepted by ACL 2026.
- 2026.01 🎉 Cave-Agent is open-sourced: an object-oriented agentic framework with superior token-efficiency, empowering the agentic AI system being built by the Hong Kong Government.
- 2025.12 🎉 We released our new work GoRL for online Reinforcement Learning with diffusion/flow-based generative models. ([Paper], [Code])
- 2025.10 🎉 OSCAR (training free technique for diverse image generation of Flow-based models) is on Arxiv. ([Paper], [Code]).
- 2025.08 🎉🎉 Received my B.Sc from The Chinese University of Hong Kong with 1-st class honor, looking forward to new journey!
- 2025.05 🎉🎉 EBC is accepted by ICML 2025 (Code available).
- 2025.03 🎉🎉 One paper accepted by ICLR 2025 (generative models for robot learning workshop).
- 2024.12 🎉🎉 One paper accepted by AAMAS 2025 (oral).
- 2024.12 🎉🎉 One paper accepted by AAAI 2025 (oral).
- 2024.09 🎉🎉 One invention patent is published.
- 2024.09 🎉🎉 Awarded Academic Performance Scholarship (for top 5% students) for consecutive two years.
- 2024.05 🎉🎉 One invention patent is officially granted.
- 2023.12 🎉🎉 As tech co-founder, I co-founded enterprise “Metasequoia Intelligence” based in Shenzhen, China.
- 2021.09 🎉🎉 Awarded Zhejiang Guolong Inspirational and Diligentia Bowen Scholarship to support my undergraduate studies in CUHK(SZ).
- 2019.09 🎉🎉 Lucky to win the 1-st prize in Provincial Chinese Mathematics Olympiad (CMO). Thanks for this intellectually-rewarding experience.
📝 Selected Publications
Conference papers and Preprints
Please scroll down to view more. * denotes joint-first-author and equal contribution.
-
Technical Report
Technical ReportCave-Agent is empowering the agentic AI system being built by the Hong Kong Government. [website]
Object-oriented Agentic Systems, extending the philosophy of Anthropic Agent-skills, superior token-efficiency. -
preprint
preprintGeneric Framework for Online Reinforcement Learning with Generative Policy class (e.g., Diffusion Policy & Flow-Matching Policy). -
ICML 2025
Forty-Second International Conference on Machine LearningOne-line code brings dynamic gradient-guided policy diversity for reinforcement learning. -
preprint
preprintEnables Flow-Matching model to generate diverse but alignment-respecting images in training-free manner. -
Preprint
PreprintDive into the creativity of diffusion language models. -
preprint
preprintReinforcing Self-Compression for Optical Agent Memory.
Invention Patents
As these works are patented in China, all these names are directly translated from Chinese.
-
A Method, System, Terminal Device, and Storage Medium for Air Quality Spatial Inference (Granted)
Inventor: Jun Song, Yibo Xu, Yiwen Pan, Maohao Ran, Zhenglin Wan, Xiaoyun Yan, Yike Guo
-
A Single-UAV Atmospheric Pollutant Source Tracing Method Based on Gradient Ascent and Physical Kinematics (Public)
Inventor: Zhenglin Wan, Jun Song, Yibo Xu, Maohao Ran, Yike Guo
White Paper
As these works are presented in China, these names are directly translated from Chinese.
- White Paper on Cross-Border Economic Large Language Model
📖 Educations
-
Doctor of Philosophy (Ph.D)
National University of Singapore (NUS)
- Affiliation: Department of Computer Science, School of Computing
- Advisor: Prof. Yang You
2026.01 -

-
Bachelor of Science (B.Sc)
The Chinese University of Hong Kong (CUHK)
- 1-st class honor
- Major: Statistics & Data Science, GPA: 3.85/4.0, Rank: 7%
- Finished my undergraduate in CUHK-Shenzhen campus, while the degree is offered by CUHK.
2021.09 - 2025.08

💻 Internships and Work Experiences
-
Full-time Research Staff
Nanyang Technological University, Singapore
- Affiliation: College of Computing and Data Science (CCDS)
- Advisor: Prof. Bo An
2025.09 - 2026.01

-
Intern (Remote)
Hong Kong Generative AI Research & Development Center (HKGAI), HKUST
- Affiliation: HKGAI
- Director: Prof. Yike Guo
- Proposed a new product-inspired Agentic Function Calling paradigm ([[CaveAgent]]).
2025.05 - 2025.09

-
Intern Researcher
Agency for Science, Technology and Research (A*STAR), Singapore
- Affiliation: Centre for Frontier AI Research (CFAR), Institute of High Performance Computing (IHPC)
- Advisor: Prof. Ivor Tsang, Dr. Xingrui Yu
2024.07 - 2025.09 (3 months full-time + 11 months part-time)

-
Research Assistant
The Chinese University of Hong Kong, Shenzhen
- Affiliation: School of Data Science
- Advisor: Prof. Jianfeng Mao
2023.06 - 2024.06

-
Machine Learning Algorithm Engineer Intern
HUIYINTONG, Shenzhen, China
- Mentor: Dr. Jun Song
2023.07 - 2024.01
🎤 Invited Talks
- Object-Oriented Agent Infrastructure — Invited by Qingke AI Community
- Stable Online Alignment of Diffusion-based Foundation Model — Invited by Meta (MRS)
🎈 Services
- Reviewer of AAAI, ICLR, ICML, NeurIPS
🎖 Honors, Awards and Scholarships
-
NUS Research Scholarship (approx. ¥250000/year plus full tuition fee subsidy, for entire Ph.D studies in NUS)
-
1-st class honor undergraduate student awarded by The Chinese University of Hong Kong
-
Yearly Academic Scholarship: B Class (for GPA Top 3%, ¥40000)
-
Yearly Academic Scholarship: C Class (for GPA Top 5%, ¥20000)
-
Yearly Dean List Award (Outstanding 1-st class Performance, for 3 years)
-
Diligentia Bowen Scholarship (¥120000, Undergraduate Admission Scholarship for 1-st prize in Provincial CMO)
-
Zhejiang Guolong Inspirational Scholarship (¥120000, Undergraduate Admission Scholarship for top 0.5% students in Chinese College Entrance Exam)
-
1st-Prize in Chinese Mathematics Olympiad (CMO)-Chongqing Province
💬 Press/Media

- The co-author of the first White Paper on Cross-Border Economic Large Language Model in Shenzhen, China. 深圳卫视:深圳发布首个跨境经济大模型白皮书
Miscellaneous
- In my spare time, I’m an music enthusiast. I’ve been playing guitar for more than 10 years and began teaching myself the piano when I was 15. During my undergraduate, I played music in two bands: “Minor Blue” and “Major Pink.” See our photos:
-
I am also a 15-years chess player, with the honor of “National Level-3 Athlete”. I love the process of comprehensive planning, logical-thinking and reasoning. Visit my Lichess profile.
-
I love play basketball 🏀. Sports makes me energetic.
-
I play video games like League of Legends, where I achieved the “diamond” level as my historically highest honor. I also play 3A games like Elden Ring, Dark Souls, Nier Automata, and elder scrolls.
-
I have a deep interest in philosophy of mind, particularly Buddhism and Taoism, as paths to explore the fundamental nature of human existence. I am also intrigued by the potential integration of these philosophical insights with modern artificial intelligence.