Qiyuan Shi

Qiyuan Shi史启源

PhD Student, Computer Science · Rutgers University–New Brunswick

About

I am a first-year PhD student in Computer Science at Rutgers University–New Brunswick, advised by Prof. Ruixiang (Ryan) Tang. I work on making large language models more controllable and more accountable: activation steering, synthetic training data, and auditing how models use their training data.

Before Rutgers, I received my B.Eng. in Computer Science from the School of the Gifted Young at the University of Science and Technology of China (USTC). I worked with Prof. Yinzhi Cao at Johns Hopkins University on data-use auditing and membership inference, and with Prof. Qi Liu at USTC on fairness.

News

Research

Beyond Single-Attribute Auditing: A Configurable Multi-Attribute Benchmark for Intersectional Fairness in LLMs
Zheng Zhang, Yifan Wu, Yuxiang Zhou, Qi Liu, Qingyang Mao, Ning Li, Weibo Gao, Qiyuan Shi, Zhenya Huang, Baosheng Yu
EMNLP 2026Main Conference
A configurable benchmark for auditing LLM fairness across combinations of demographic attributes, rather than one attribute at a time.
DAudit: Data-use Auditing of Large Language Models
Under reviewJHU · advised by Yinzhi Cao
Black-box auditing of whether specific data was used to post-train an LLM, under label-only or logits access, with formal false-discovery-rate guarantees.
Membership Attribution against Large Language Models
Under reviewJHU · advised by Yinzhi Cao
Extends membership inference from “was this sample used?” to “in which training stage was it used?”, under low false-positive constraints and restricted logit access.
UpdateSteer: Set-Conditioned Activation Control for Synthetic Training Data
In preparationRutgers · advised by Ruixiang Tang
Jointly chooses activation interventions for a whole set of synthetic examples, using learner gradients and set-level matching so that each example adds a complementary training signal.

Education

Experience

Honors