CRiSM Seminar - Sequential learning via a combined reinforcement learning and data assimilation ansatz for decision support
Speaker: Jana De Wiljes
Abstract: In many applicational areas there is a need to determine a control variable that optimizes a prespecified objective. This problem is particularly challenging when knowledge on the underlying dynamics is subject to various sources of uncertainty and access to observations is severely limited. A scenario such as that for example arises in the context of therapy individualization to improve the efficacy and safety of medical treatment. Mathematical models describing the pharmacokinetics and pharmacodynamics of a drug together with data on associated biomarkers can be leveraged to support decision-making by predicting therapy outcomes. We present a continuous learning strategy that allows to sequentially update the model parameters and states via a particle-based data assimilation scheme and combine it with reinforcement learning to tailor the dosing policy to the specific patient. The proposed technique is applicable to various other application areas and we explore how the underlying uncertainties reflect in the approximated control variable.