This project sits with the ACRC Academy , a dedicated centre for doctoral training based with the Advanced Care Research Centre. The programme is novel, structured, thematic, cohort-based, and of 48 months duration, with an initial taught year followed by three years of PhD research. Each PhD research project within the Academy has been devised by a supervisory team comprising academic staff from at least two of the three colleges within the University of Edinburgh.

Project  

The ultimate decision for a carer is to predict an intervention’s outcome e.g. medication, diet etc. An elegant framework for such tasks, is causal machine learning [1]. This project will start by integrating data being made available by the ACRC for an exemplar caring decision and use a simple causal predictive model to develop a demo application. As simple models cannot scale as the number of information sources increase, non-linear causal models will then be developed [2].  This will require causal structure discovery: finding useful variables and their causal associations. To address this, we will combine representation learning and causality [3].  A key desire for any AI is to be fair and transparent. While causal models by definition should be explainable, we will study whether predictive models based on causality do reduce risks of bias and increase fairness and transparency.   

The student will benefit by expertise in machine learning and healthcare (Tsaftaris) and application of data-science in population health (Harrison). The student will work together with the ACRC and the Tsaftaris group. In addition, Tsaftaris works closely with Canon Medical Research Europe [CMRE] and their Center of Excellent in AI and the student can and will interact with CMRE scientists.  Our weekly discussion group (and journal club) is attended by members across the VIOS collaboratory (of Tsaftaris, https://vios.science/), Canon scientists and other collaborators. 

Causal inference, machine learning, fairness, predictive modeling 

1. Bernhard Schölkopf, Causality for Machine Learning, https://arxiv.org/abs/1911.10500 

2. Louizos et al, Causal Effect Inference with Deep Latent-Variable Models, NIPS 2017 https://arxiv.org/abs/1705.08821  

3. Schölkopf et al, Towards Causal Representation Learning, Proceedings of the IEEE, Special issue on Advances in Machine Learning and Deep Neural Networks, 2021.