Methods from data assimilation, inverse problems, and machine learning have shown exciting potential for transforming biomedicine.
First, I will show how a Bayesian Inverse approach can be used to build patient-level phenotypes of diabetes, despite noisy data conditions and weakly identifiable models. Then, I will discuss an operationalized application of data assimilation to real-time, personalized forecasting of blood glucose levels for people with type 2 diabetes. Finally, I will point to inadequacies of these approaches that stem from model error, and present novel machine learning approaches that can account for arbitrary errors in modeling and data assimilation.
Invited by Jana de Wiljes