Research Directions
We develop adaptive AI systems that learn from data and decide what to do next (where to collect data, which intervention to try, how to allocate limited resources) in complex scientific and public health settings. Our work combines sequential decision-making (reinforcement learning, Bayesian optimisation and active learning) with probabilistic modelling. We focus on methods that stay reliable when feedback is limited, indirect or changing over time.
🩺 AI for Health: Sequential Decision Making for Infectious Disease Forecasting and Surveillance
Public health decisions are made step by step and under uncertainty: which regions to monitor, where to test, and how to allocate scarce resources as an outbreak unfolds. We design algorithms that use forecasts, and their uncertainty, to guide these choices, for example graph-based active learning for disease surveillance (PNAS 2024). See also our review of artificial intelligence for modelling infectious disease epidemics (Nature 2025).
- 🌍 Spatiotemporal forecasting and resource allocation for healthcare and epidemic response
- 📈 Nowcasting and forecasting of epidemic trajectories from dynamic, often delayed data
- 🔄 Non-stationary decision making when spatial–temporal signals change over time
- 🧩 Sequential disease surveillance and control through reinforcement learning
🤖 Foundation Models for Health and Adaptive Inference
Large pretrained models offer a new way to forecast and support decisions across many diseases and regions at once. Rather than building them from scratch, we use and improve foundation models for health data, and study how they can be adapted at test time (fine-tuned or updated as new data arrive) so they stay robust under data drift and uncertainty. Related work includes scalable spatiotemporal inference with transformer neural processes (AISTATS 2026).
- 🧠 Using and improving foundation models for spatial–temporal health forecasting and intervention planning
- ⚙️ Test-time adaptive inference and fine-tuning for rapidly evolving data environments
Reaching Out
If you are interested in the research directions above, please email me (mengyan.zhang@bristol.ac.uk) with [PhD Application + Your name] in the subject line. You will need:
- A CV;
- A Personal Statement, which is a one- to two-page document introducing yourself and outlining your motivation for PhD research;
- A transcript of any qualifying degrees (completed and/or underway);
- A research proposal (optional, but preferable);
- Any additional materials to support your application, e.g. research outputs, thesis, coding repository, etc.
Due to the volume of enquiries, only shortlisted candidates may receive a response.
PhD Opportunities
Fully funded PhD opportunities are available (deadlines checked October 2026; please confirm on each website):
- University of Bristol Science & Engineering PhD scholarships: deadline expected January 2027 (not yet announced; the 2026/27 round closed on 19 January 2026).
- China Scholarship Council-University of Bristol PhD Scholarship (CSC): submit your PhD application, with a valid English language certificate, by 12 noon (UK time), 30 November 2026.
- Centres for Doctoral Training (CDTs), e.g. UKRI CDT in Practice-Oriented Artificial Intelligence, September 2027 entry: 30 October 2026 (1st round), 26 February 2027 (2nd round), 14 May 2027 (final round).
I am also willing to be a secondary supervisor for PhD students in related areas.