Uncertainty quantification for multi-physics simulations

Multi-physics or multi-component systems are complex systems made of collections of nested systems with properties such as i) similar individual components working at different time and spatial scales; ii) with a potentially mammoth number of components; or iii) with components non-linearly connected or interacting in a network. Examples of such systems include the composite design of materials (components operating at different scales), carbon sequestration (components described by different

models), and the electric power grid (individual components acting in a network).

Simulation of multi-physics systems is computationally expensive. Although computer emulation is a relatively mature technology that reduces simulation time and addresses questions such as uncertainty quantification, sensitivity analysis, and calibration of individual components, computer emulation technologies for multi-physics simulations are in their infancy and have scarcely been developed both at a theoretical level and as practical tools.

In this proposal, we are concerned with individual computer emulators that can be composed in hierarchical structures. In the machine learning literature, these models are sometimes known as deep Gaussian processes (deep GPs), whereas in the computer emulation literature, they are known as linked emulators. Most emphasis in the ML literature has been on providing tractable inference approaches, with deep GP models mostly used for prediction tasks. We aim to bridge the gap between the generic model formulation of the composition of computer emulators, i.e., deep GPs, and the more challenging real-world scenarios of multi-physics simulations starting with an application to fusion energy.

Desirable Student Background:

A student with an MSc in Physics, Computer Science, Statistics, Mathematics or related field is the ideal candidate for this project. They don’t need to have previous knowledge of fusion energy. This PhD project will emphasise methods development.

How to apply:

Student should submit an application through the below link under the PhD in Artificial Intelligence CDT.