Academic:

I am a postdoctoral researcher at Oxford Statistics working with Prof. Patrick Rebeschini. My research aims to develop statistical and information-theoretic foundations that make machine learning systems more efficient and reliable, especially when data is heterogeneous, high-dimensional, or changing over time.

Empirical progress in machine learning continues to outpace our theoretical understanding of it. I work on the slow side of that gap: characterising what is achievable in principle, and designing systems and algorithms that approach those limits.

I have primarily been working on three problems recently:

• Mathematically understand common heuristics and phenomena in model training, such as early stopping of gradient descent, benign overfitting, and feature learning
• Sequential decision-making on heterogeneous data with rigorous guarantees, including in bandits and adaptive inference
• Uncertainty quantification under minimal assumptions, using e-values, conformal prediction, and prediction-powered inference

I completed my PhD at Cambridge, where I was fortunate to be advised by Prof. Ramji Venkataramanan. My PhD work centered on information theory and high-dimensional statistics. Prior to my PhD, I obtained my BA and MEng in Information and Computer Engineering from Cambridge.

I was an intern with Microsoft Research Cambridge in 2023, where I designed error correction codes for Project Silica, a novel glass-based archival data storage technology. This work has recently appeared in Nature.

Personal:

My first name is pronounced “Sh-ih-ow Chi” (meaning jade born at dawn in Chinese), but feel free to call me Shirley! I publish under both Xiaoqi Liu and Shirley Xiaoqi Liu.

Living and working across British, Chinese, Singaporean, and American cultures has shaped how I engage with people and ideas. I care about the communities I’m part of and have taken on roles like postgraduate representative to help bring people together.

Get in touch:

I am actively seeking collaborators on machine learning theory and statistics. Please send me an email if you are interested in working together! I’m always happy to chat.

Email: shirley.liu [at] stats.ox.ac.uk, shirley.xiaoqi.liu [at] gmail.com