01 / EFFICIENT ADAPTATIONICML 2024
Less to learn.
More to adapt.
More to adapt.
FourierFT / C³A
Rethinking fine-tuning
PEFT · FOURIER METHODSAdapting foundation models with compact spectral representations and circular convolution.
RESEARCHER & ENGINEERHKUST (GUANGZHOU)
AOCHUANCHEN
I study how to make AI
more efficient. More useful.
PhD in Computer Science
From optimization to practical AI systems.
SELECTED
RESEARCH.
01 — IDEAS INTO SYSTEMS02 / GRADIENT-FREE LEARNINGICLR 2024
DeepZero
f(x + μu) − f(x − μu)2μ
Learning beyond backpropagation
Training without gradients
ZEROTH-ORDER OPTIMIZATIONScaling zeroth-order optimization to deep model training, with applications to black-box scientific problems.
03 / VISUAL PROMPTINGCVPR 2023
INPUT
+ PROMPT
+ PROMPT
FROZEN
MODEL
MODEL
LABEL
MAPPING
MAPPING
A new task.
The same model.
The same model.
Understanding what makes prompts work
A different way to transfer
VISION · TRANSFER LEARNINGUnderstanding and improving visual prompting through the relationship between prompts and label mappings.
04 / PRACTICAL AI SYSTEMSRESEARCH → DEPLOYMENT
Ideas.
Code.
Impact.
Code.
Impact.
DESIGNGENERATEEVALUATE
Agents & applied machine learning
Building AI that works
AGENTS · APPLIED AIDesign-centric coding agents, automated evaluation, and machine learning for real-world payment and gaming systems.
BEHIND
THE WORK.
02 — ABOUT MEAOCHUAN CHEN
PhD student / Researcher
PhD student / Researcher
My work connects
optimization, efficient learning,
and practical AI systems.
I am a PhD student in Computer Science at the Hong Kong University of Science and Technology (Guangzhou), advised by Dr. Jia Li and Dr. Xiaowen Chu. Before that, I received my bachelor's degree in Automation from Tsinghua University.
My research spans visual prompting, parameter-efficient fine-tuning, and zeroth-order optimization. I also build AI systems, with a particular interest in turning algorithmic ideas into useful tools.
PAPERS &
PERSPECTIVES.
Google Scholar ↗More publications
03 — FROM ALGORITHMS TO DEPLOYMENT
RESEARCH,
IN PRACTICE.
Design-centric coding agents
At Tencent, I built an end-to-end software engineering agent workflow, developed automated benchmark case generation, and evaluated design agents’ coding capabilities on ProgramBench.
Machine learning at scale
My industry collaborations include mini-game lifetime value prediction in WeChat and merchant category identification in Weixin Pay, with work published at KDD 2025 and KDD 2026.
THE
JOURNEY.
04 — EXPERIENCE & EDUCATION2025 — PRESENT
Tencent
Intern · Qingyun Project
Design-centric software engineering agents and evaluation.
2024 — PRESENT
HKUST (Guangzhou)
PhD · Computer Science
Advised by Dr. Jia Li and Dr. Xiaowen Chu.
2024
IDEA
Research Intern
International Digital Economy Academy, Shenzhen.
2023Lawrence Livermore
Lawrence Livermore
National Laboratory
Research Intern
Zeroth-order optimization for physics simulation.
2018 — 2022
Tsinghua University
BEng · Automation
Department of Automation, Beijing.