Bio
Kaitlyn Louth is a PhD student at The University of Edinburgh and Heriot-Watt University, within The Maxwell Institute's Centre for Doctoral Training Programme in the Schools of Mathematics. Her research lies within prediction risk modelling, more specifically using Bayesian multilevel modelling and Bayesian neural networks to predict critical illness morbidity risk. She is also the co-lead for The Piscopia Initiative, which is a UK-wide programme which aims to encourage more women and underrepresented genders to pursue a PhD in mathematical sciences and create a sense of community for those already undertaking a PhD.
Research interests
Kaitlyn’s project aims to build, assess and improve prediction models for analysing morbidity risk for critical illnesses. Morbidity risk is an individual’s likelihood of being diagnosed with a specific critical illness. These include serious conditions such as heart attacks, strokes and multiple sclerosis, and diseases including cancers and Parkinson’s disease. She is improving on existing models by combining traditional statistical methods, namely hierarchical Bayesian models, with artificial neural network deep learning techniques. This hybrid approach therefore explores how to use Bayesian Neural Networks (BNNs) to build morbidity risk prediction models. In BNNs, all weights and biases in the neural network have a probability distribution attached to it, allowing us to learn a probability distribution over all neural networks. This helps to mitigate overfitting and importantly provides information on uncertainty quantification. This hopes to be a vast improvement on the standard black-box artificial neural networks, whose outputs do not necessarily provide information on the structure of the network or on the estimation of uncertainty of predictions. These models are data-driven, using sources from the Office for National Statistics (ONS), NHS Open Data platform, Public Health Scotland, and insured population data from the Institute and Faculty of Actuaries (IFoA).