Ziping Xu 
Postdoctoral Fellow / Department of Statistics / Harvard University
I am a Postdoctoral Fellow in Statistics at Harvard University with Professor Susan Murphy.
I obtained my PhD degree in Statistics from University of Michigan advised by Professor Ambuj Tewari. Before joining University of Michigan, I obtained my B.S. degree in Data Science from Peking University in 2018, where I was advised by Professor Song Xi Chen.
Find my lastest CV here.
Research Interests
My primary research interests lie in machine learning theory and applications, with a focus on sequential decision-making approaches for digital interventions in domains such as mobile health. My theoretical works tackle the critical challenges in applying sequential decision-making approaches, for example, Reinforcement Learning (RL), in mobile health. Below I summarize three key challenges and my works on them:
- Sample-Efficient Online Algorithm Design: In mobile health, the availability of data is often limited, making sample efficiency crucial. I develop algorithms that leverage structural information about the domain to learn effective policies with small sample size [12,15,19].
- Statistical Inference for Adaptively Collected Data: Adaptive experimental designs, commonly used in mobile health, introduce dependencies across time and users, violating the assumptions of most existing statistical inference methods. I develop statistical inference approaches that remain valid in these challenging settings [1,7].
- Transfer Learning/Multitask Learning: Due to limited sample sizes in mobile health, transferring knowledge from existing datasets is often necessary. However, significant domain shifts, driven by differences in participant demographics, intervention designs, or societal changes, complicate this process. I design transfer learning approaches that account for such shifts, enabling more robust and effective learning [4,5,6,11,17].
Mobile health clinical trial design: I am actively involved in designing mobile health clinical trials with domain scientists. One such trial is ADAPTS-HCT, which targets medication adherence among adolescents and young adults (AYA) undergone bone marrow transplantation. I lead the RL algorithm design for adaptive delivery of digital interventions as part of the intervention package. See my talk at INFORMS 2024 for more details.
Publications (Google Scholar)
Recent Invited Talks
- AI and Healthcare Seminar Series at MILA, November 2024
- STATS 300 Seminar at Department of Statistics, Harvard University, Boston, MA, November 2024
Designing a Multi-agent RL Algorithm for Improving Post-HCT Medication Adherence via a Digital Intervention
- 2024 INFORMS Annual Meeting, Seattle, WA, October 2024
The Fallacy of Minimizing Cumulative Regret in the Sequential Task Setting
- 2024 Joint Statistical Meetings, Portland, OR, August 2024