Recent pandemics such as COVID-19 exposed major weaknesses in current epidemic monitoring and forecasting models. Public health decision-making is often hindered by under-reporting of infections, delays and noise in epidemiological data, and rapidly changing transmission dynamics driven by behavioural, biological and policy factors. Traditional epidemiological models struggle to adapt to these changes, while purely data-driven machine learning approaches often lack the biological realism required for reliable policy support.
This 3-years, fully funded PhD project aims to address these challenges by developing a Physics-Informed Digital Twin (PIDT) for epidemic intelligence. The project will combine mechanistic epidemiological models with modern machine learning techniques to create a continuously updating computational model capable of monitoring epidemic dynamics, estimating hidden infection levels, and simulating intervention strategies.
The core idea is to integrate epidemiological compartmental models (such as SIR or SEIR models) with Physics-Informed Neural Networks (PINNs). PINNs are a class of machine learning models that embed physical or biological governing equations directly into the learning process. By incorporating epidemiological knowledge from biology and medicine into the training of neural networks, the resulting models maintain interpretability and realism while remaining flexible enough to adapt to noisy and incomplete real-world data.
Within this framework, the PhD researcher will develop methods capable of inferring time-dependent transmission parameters, hidden infection dynamics, and levels of under-reporting from incomplete epidemiological data. Bayesian uncertainty quantification techniques will be incorporated to ensure that the resulting predictions and simulations provide robust estimates together with confidence measures.
A key component of the project is the development of a Digital Twin framework for epidemic systems. Digital Twins are dynamic virtual models that continuously update as new data become available. In this project, the Digital Twin will ingest streaming epidemic data, update model parameters in real time, and simulate alternative scenarios of disease spread. These simulations will enable the evaluation of potential public-health interventions such as vaccination campaigns, social distancing measures, or healthcare capacity constraints.
To support policy analysis, the project will also integrate optimisation methods that allow intervention strategies to be explored within realistic social and economic constraints. Multi-objective and stochastic optimisation approaches will be used to study trade-offs between competing objectives such as reducing infection levels, limiting hospital stress, and minimising economic disruption.
The project sits at the intersection of applied mathematics, machine learning, epidemiology, and optimisation, and will contribute to the development of interpretable AI tools for epidemic preparedness and response. Expected outcomes include new methods for physics-informed machine learning in epidemiology, improved techniques for handling under-reported epidemic data, and an open-source software framework for epidemic Digital Twins.
The research will contribute to emerging efforts to develop AI-driven decision-support systems for public health, with potential impact for organisations such as public health agencies, healthcare systems, and government planning bodies.
