Risk CDT - Autonomous Navigation of the Formulation Space

PLEASE APPLY ONLINE TO THE SCHOOL OF ENGINEERING, PROVIDING THE TITLE AND NAME OF THE PRIMARY SUPERVISOR

This is a project within the multi-disciplinary EPSRC and ESRC Centre for Doctoral Training (CDT) on Quantification and Management of Risk & Uncertainty in Complex Systems & Environments, within the Institute for Risk and Uncertainty. The studentship is granted for 4 years and includes, in the first year, a Master in Decision Making under Risk & Uncertainty. The project includes extensive collaboration with prime industry to build an optimal basis for employability.

Formulation is the task of deciding on the proportions of ingredients that are used to make mass-market products (eg washing powder and shampoo). The merits (e.g. colour, effectiveness etc) of a given set of proportions (and various other parameters of the mixing process) can be evaluated by a scientist making a potential new product and measuring it. However, even with a relatively small number of ingredients, it rapidly becomes infeasible to fully explore all possible formulations and measure each formulation’s merits: the number of possible formulations is just too big. This can be offset by measuring multiple formulations at once and by using computer simulations that can, to some degree, predict the measurements that are likely. However, current practice is inherently risk averse and involves scientists making small perturbations about formulations that are known empirically to work well. This project relates to posing this problem as one involving autonomously deciding on the formulations to measure. The aim is to learn about the uncertain mapping of formulation to measured merits while simultaneously identifying the optimal formulation. Problems of this type are well understood in the mathematical guise of Partially Observed Markov Decision Processes (POMDPs). While POMDPs have been used to combine exploration and exploitation in other contexts (e.g. robotics), its use in the autonomous iterative definition of (physical and simulated) experiments is new. The work builds on strong ongoing interactions between the University of Liverpool and Unilever, who are very keen to migrate towards autonomous formulation, e.g. as part of their involvement in the University of Liverpool’s £68M Materials Innovation Factory. The successful applicant will end up having skills in statistics, computing and chemistry. Starting with a strong background in a subset of these disciplines with the enthusiasm to learn about any others is therefore strongly desirable.