Qiuyi Ding

I am an undergraduate student in Computer Science at the University of Michigan. I'm advised by Prof. Andrew Owens during my honors program. I'm also fortunate to collaborate with Prof. Ang Chen and Patrick Kon, who provided me with invaluable research experience.

Email / Scholar / Github

Research Interests

My research interests lie in controllable generative models and multimodal perception systems.

News

01/2026: 🎉 One paper accepted to ICLR 2026, See you in Brazil!

Publications

HidingSound: Adaptive Soundscape Generation for Online Noise Coverage
Qiuyi Ding, Yiming Dou, Andrew Owens
2026

An online soundscape generation framework that perceptually covers urban noise with adaptive natural soundscapes, enabling more comfortable listening experiences in real-world environments such as New York City.

Curie: Toward Rigorous and Automated Scientific Experimentation with AI Agents
Patrick Tser Jern Kon^*, Jiachen Liu^*, Qiuyi Ding, Yiming Qiu, Zhenning Yang, Yibo Huang, Jayanth Srinivasa, Myungjin Lee, Mosharaf Chowdhury, Ang Chen
arXiv preprint 2025
project page / arXiv / GitHub

An AI agent framework designed to automate scientific experimentation with rigor at its core. From reproducing existing results to extending and challenging them, Curie enables researchers to move scientific research at the speed of thought.

EXP-Bench: Can AI Conduct AI Research Experiments?
Patrick Tser Jern Kon^*, Qiuyi Ding^*, Jiachen Liu^*, Xinyi Zhu, Jingjia Peng, Jiarong Xing, Yibo Huang, Yiming Qiu, Jayanth Srinivasa, Myungjin Lee, Mosharaf Chowdhury, Matei Zaharia, Ang Chen
ICLR 2026
arXiv / Hugging Face / GitHub

We introduce a large-scale benchmark of 461 AI research experiments and 12,000+ subtasks to evaluate whether AI agents can conduct end-to-end scientific experimentation.

Selected Projects

Immersive NPCs in VR Game Environment (2024)
Course Project for EECS498: AI-Enabled Mixed Reality

We built LLM-powered NPCs in Unity + VRChat, enabling natural conversations, personality-driven behaviors, and memory systems that let NPCs remember different users
By combining speech-to-text, text-to-speech, and real-time reasoning with GPT-4, we created more immersive, believable, and engaging VR gameplay experiences

Encoder Integration and Efficiency Optimization for Language-Embedded Radiance Fields (2024)
Course Project for EECS442: Computer Vision

Integrated a custom SigLIP encoder into the LERF pipeline, achieving higher accuracy in open-vocabulary 3D semantic localization
Conducted efficiency optimization in the LERF training pipeline by introducing AdamW in place of Adam/RAdam and reducing evaluation/checkpointing overhead, achieving a 31% improvement in training efficiency

Misc.

I enjoy playing tabletop role-playing games such as D&D and COC. I've been fortunate to meet some wonderful friends through in-person sessions. Here's a photo of our dice collection.

Website template from Jon Barron.
Last updated Jul 2025.