We are seeking an exceptional PhD candidate to work across the fields of reinforcement learning and health services. In reinforcement learning an artificially intelligent ‘agent’ acts as an optimiser and interacts with an uncertain environment in order to obtain ‘reward’. Unlike supervised learning where models are trained on large datasets, an agent starts with no data and must be trained in a synthetic environment that mimics a real system. There are many ways to train an agent including, self play where an agent plays against historical versions of itself to improve. Many applications of reinforcement learning have been in the field of games. For example, training reinforcement learning agents to beat the world’s top players in Go, Chess and the video game Starcraft 2. These highly constrained settings have provided ideal opportunities to study reinforcement learning, but there has been little research tackling the challenges an agent would face in real world systems which are far more complex and unconstrained. Reinforcement learning offers substantial potential for future health care in terms of efficiency and patient outcomes. This is an exciting research area filled with many method and application challenges including capturing the diversity of a real world health system in a training environment, and agent reward gaming.
This doctoral project will work in the methodological area of reinforcement learning and in particular how to train competent agents to manage health service delivery systems such as ambulance services, emergency departments and elective and emergency operations. A successful candidate will initially focus on developing computer simulation models, such as discrete event and agent based simulations, from the health services and operational research literature, and adapting them to generate data to train standard reinforcement learning agents such as those provided by Open AI. A specific focus of the work will be on use of neural network architectures, but there will be opportunity for classical meta-heuristic and evolutionary algorithms to serve as comparators in the optimisation of models. The candidate will also explore novel approaches to training high quality agents, the impact of different reward structures and new architectures.
This is unique opportunity for a talented individual to work in one of the most important areas of modern AI. The successful candidate will join PenCHORD: a thriving data science and operational research team in Exeter’s Medical School. The team has had successes in applying data science and mathematical modelling to many aspects of health care including diabetes, cancer, acute stroke, mental health, ambulance services, emergency departments and COVID19. PenCHORD is funded by the National Institute for Health Research (NIHR) Applied Research Collaboration South West Peninsula (also known as PenARC). PenARC is one of 15 ARCs across England, part of an £135 million investment by the NIHR to improve the health and care of patients and the public. Upon joining the team a candidate will be trained in modern approaches to open science, following the Turing Way, covering skills to provide runnable and transparent models and artefacts for other researchers and the NHS. Training in machine learning and computer simulation will be provided as required.