I received a GAGE Fellowship for Spring 2027 from the Department of Computer Science & Engineering at the University of Minnesota.
PhD Student in Computer Science
Shuyu Gan 甘书宇
Pronounced: SHOO-yoo GAHN
University of Minnesota, Twin Cities
I am a Computer Science PhD student at the University of Minnesota, where I work with Minnesota NLP and the Optimization for AI Lab. I am co-advised by Prof. Dongyeop Kang and Prof. Mingyi Hong.
My research interests include autoresearch, test-time reasoning, and GUI agents. My long-term goal is to develop general-purpose agents that can reliably complete complex computer tasks.
I received my bachelor's degree in Computer Science and Technology from the Turing Class at Zhejiang University in 2024.
Autoresearch
Can agents train and audit models on their own?
Research Interests
-
Autoresearch
Agents that train models and audit training runs on their own.
-
Test-Time Reasoning
Spending inference-time compute to help models reason better.
-
GUI Agents
Multimodal agents that operate computers through graphical interfaces.
News
We introduce SanSi, a looped typed decision model for System 1.5 thinking. Check out our paper!
Do LLMs Recognize Your Latent Preferences? was accepted to COLM 2026.
Scaling Unverifiable Rewards was accepted to ACL 2026 Findings.
A2P-Vis received an Honorable Mention at the VISxGenAI Workshop at IEEE VIS 2025.
Agent S was accepted to ICLR 2025.
I graduated from Zhejiang University and was nominated as an Outstanding Graduate of Zhejiang University.
Research Experience
Industry
Research Intern
Applied Sciences Group, Microsoft
Mentored by Yinheng Li, studying whether AI agents can autonomously supervise model training.
Research Intern
Zhipu AI
Mentored by Xiao Liu, researching tool-based and agentic reinforcement learning for computer-use agents.
Research Intern
Simular
Worked with Prof. Xin Eric Wang on Agent S, an MLLM-based framework for computer-use agents.
Academia
Graduate Research Assistant
Minnesota NLP and OptimAI Lab, University of Minnesota
Agentic AI and inference-time reasoning, advised by Prof. Dongyeop Kang and Prof. Mingyi Hong.
Undergraduate Visiting Research Intern
mlPC Lab, UC San Diego
Studied concept-guided diffusion training with Prof. Zhuowen Tu.
Research Intern
IDEA Lab, Zhejiang University
Worked on culture-aware multimodal co-speech gesture generation with Prof. Lingyun Sun.
Selected Publications
Google ScholarSanSi: A Looped Typed Decision Model for System 1.5 Thinking
Shuyu Gan, Y.-J. Lee, D. Kang
Featured by Turing Post
PaperScaling Unverifiable Rewards: A Case Study on Visual Insights
Shuyu Gan, J. Mooney, P. Hao, R. Wang, M. Hong, Q. Wang, D. Kang
PaperDo LLMs Recognize Your Latent Preferences? A Benchmark for Latent Information Discovery in Personalized Interaction
I. Tsaknakis, B. Song, Shuyu Gan, D. Kang, A. Garcia, G. Liu, C. Fleming, M. Hong
PaperA2P-Vis: An Analyzer-to-Presenter Agentic Pipeline for Visual Insights Generation and Reporting
Shuyu Gan, R. Wang, J. Mooney, D. Kang
Honorable Mention
PaperAgent S: An Open Agentic Framework that Uses Computers Like a Human
S. Agashe, J. Han, Shuyu Gan, J. Yang, A. Li, X. E. Wang
PaperCultural Self-Adaptive Multimodal Gesture Generation Based on Multiple Culture Gesture Dataset
J. Wu, S. Chen, Shuyu Gan, W. Li, C. Yang, L. Sun
PaperAcademic Service
Reviewer
ACL Rolling Review (ARR), October 2026
Reviewer
NeurIPS 2026, ED Track
Reviewer
ICML 2026
Reviewer
ACL 2026 Workshop
Personal
Outside research, I enjoy playing CS2 and Valorant, distance running, and badminton. I was a member of Zhejiang University's Kuafu Long Distance Alliance for three years and have participated in the 1500m race and half marathon.
For research discussions or collaboration, please contact me at gan00067@umn.edu.