Yu Zhang

I develop data-efficient machine learning and digital-twin methods for engineering and healthcare. My work combines probabilistic modelling, Bayesian optimisation, active learning, uncertainty quantification and interpretable models to support decisions when experiments or observations are limited.

At Surrey, my PhD focuses on AI-enabled formulation design and drug delivery. Alongside my doctoral research, I work with Dr Xilu Wang as a Research Assistant on real-world clinical data in assisted reproduction (IVF), developing reproducible workflows for outcome modelling and patient-aware, time-aware validation.

Formulation research programme

Current research directions

• Clinical machine learning - defining outcomes, constructing reliable features and developing patient-aware, time-aware validation for assisted-reproduction data. Research overview.
• Adaptive formulation optimisation - allocating small experimental batches between performance-seeking and information-seeking candidates.
• Early experimental decision-making - forecasting final outcomes from partial IVPT/IVRT trajectories and controlling false stops.
• Interpretable dynamic modelling - embedding physical and mathematical constraints into symbolic regression for drug-release kinetics.
• Scientific LLM systems - extracting traceable formulation and permeation evidence from literature and building reusable research datasets.

Background

I am pursuing a PhD in Chemical and Process Engineering at the University of Surrey, supervised by Professor Tao Chen and Dr Dimitrios Tsaoulidis. Expected thesis submission: late 2026 / early 2027.

My research experience includes placements at Jagiellonian University, Poland (June-August 2026), supported by COST STSM, the UK Turing Scheme and FEPS, and at West China Hospital, Sichuan University (January-March 2025).

Before Surrey, I completed an MEng in Control Science and Engineering at China University of Petroleum - Beijing and a BEng in Automation at Panzhihua University. My earlier work spans industrial process monitoring, fault diagnosis, data-driven analysis of reservoir-simulation outputs and energy-market software. This background informs my focus on reproducible models that remain useful under real experimental and data constraints.