Simulating a day in a patient’s life: why and how (initial explorations)

Join us for a seminar with guest speaker Professor Paolo Missier.

Guest Speaker: Professor Paolo Missier, School of Computing, Newcastle University

Abstract: Synthetic data refers to plausible, ad hoc data points that are generated from given ground datasets. There is increasing evidence that the ability to generate synthetic datasets out of real ones may facilitate applying ML algorithms in important settings, including when the naturally available training sets are too small, or when they are biased in some regions of the features space. This is particularly important in medicine and healthcare applications, where the positive class of interest is often under-represented. Being able to generate additional "case" data points with sufficient fidelity to be usable in practice for learning is therefore important, and technically challenging. 

In this talk we are going to present initial experiments aimed specifically at replicating accelerometry traces from ground truth samples taken from UK Biobank. We then want to use those traces to simulate “a day in life” of individuals with specific, under-represented characteristics, such as a chronic condition that entails physical impairment. If successful, the approach can potentially be applied to the training of AI agents that are designed to understand people’s mobility patterns and need to learn to suggest suitable lifestyle interventions.


Join Zoom Meeting: