Uncertainty quantification for synthetic brain network models

Mathematical models are widely used in simulating and analysing the brain’s connectivity and dynamics. Within these models, coupled Hopf oscillators are employed to describe the state of a brain network model changes over time and account for the interactions between brain regions. However, the brain model also evolves in response to learning and ageing, leading to the connections between neurons being linked or disconnected over time. When these connections change, a new brain model may have different properties and behaviours than the previous one. We need to rerun the system for each new brain model, which is computationally expensive. Gaussian Process Emulators provide a cheap and efficient framework to approximate the simulator and produce the prediction of simulator output with associated uncertainty. We extend the traditional static Gaussian Process Emulator to a dynamic version, and this is used to predict the dynamic evolution of the synthetic brain network model and quantify its uncertainties.