Our vision is to establish the computational foundations for an „AI clinician“, an AI system informed by millions of patient records that is using Reinforcement Learning (RL) to learn optimal treatment policies for critically ill patients (Komorowksi et Faisal, 2018, Nature Medicine; Gottesman et al, 2019, Nature Medicine). Prof Aldo Faisal, has been awarded a prestigious UKRI Turing AI Fellowship in Reinforcement Learning for Healthcare for which are recruiting up to 3 post-doctoral Research Associates.

Our machine learning research goal is centred on developing what is needed as a theoretical foundation for taking our „AI clinician“ proof-of-principle all the way to clinical deployment, and expanding its capabilities and responsibilities. Our Research Associates doing fundamental work will be supported by a team of experienced Researchers and a program grant to take the AI Clinician through the regulatory and technical challenges of clinical deployment. The AI Clinician programme is grounded in the established foundations of clinical decision support systems, which will run as a bed-side recommender system watching over the patients state second-by-second and prompting specific clinical interventions to undertake as the patient state evolves.

We are uniquely placed to take our machine learning developments into a meaningful clinical setting, as we have been awarded with generous funding by the UK’s Ministry of Health/NHS-X to setup a globally unique AI for Healthcare testbed for reinforcement learning interventions in 4 intensive care units across Central London. On the clinical side of the AI Clinician we are collaborating with world-leading intensive care clinicians (Anthony Gordon & Matthieu Komorowksi) also at Imperial College. Our project has the support of regulatory experts and specialised medical device units that support us in taking our entire research program all the way to clinical deployment in a live hospital environment.

The overall research environment is corresponding to our ambition, the Departments where Aldo Faisal’s research group is based has been consistently ranked top 3 in the UK, and Imperial College has been consistently ranked in the top-10 universities world-wide. Imperial College boasts a world-leading ecosystem of AI for Healthcare research across the breadth of both AI and Medicine, and hosts among other things the pioneering UKRI Centre in AI for Healthcare (directed by Aldo Faisal), as well as the AI network, comprising over 200 faculty and 1000 Research Associates and PhD Students at Imperial that work in AI.

Duties and responsibilities
The aim of the project is the advancement of foundational Reinforcement Learning methods with the immediate application to clinical intervention and involves the unique opportunity to develop core machine learning theory and see it evaluated with clinical end-users as needed.

We are therefore searching for our UKRI Turing AI research programme areas at the interface of:

1. Causal Inference & Reinforcement Learning
Despite the parallels between optimal sequential decision making in Causal Inference (Dynamic Treatment Regimens) and continuous control in RL, there has only been some initial work linking these (Barenboim & Pearl, 2016; Johnson et al 2017; Gottesmann et Faisal…, 2019, Nature Med) and we are keen to synthesize these into a common theory.

2. Foundations of Reinforcement Learning
Advancing RL methods to cope with partial observability, risk, sparseness or the size of vast data sets, (including Distributional RL & Fusing Off-policy & On-Policy learning; Off-Policy Learning in POMDPs; Neural ODEs for RL), e.g. see Li & Faisal, 2021, AAAI;

3. Explainable AI & Cognitive Neuroscience
We believe that Explainability in AI requires us to develop empirical methods for measuring how well humans were explained and how much trust this explanation has caused, establishing a clear link to cognitive neuroscience. E.g. see Beyret, Shafti & Faisal, IROS 2019, Shafti et Faisal, IROS 2020.

Dear All