2020/10/13 by Taku Yamagata, Yamagata, Taku, Aisling Ann O’Kane +13
Biochemistry, Genetics and Molecular Biology · Computer Science · Medicine · #Data Stream Mining Techniques #Diabetes Management and Research #FOS: Computer and information sciences #Machine Learning (cs.LG) #Neural Networks and Reservoir Computing #Receptor Mechanisms and Signaling
paper · pdf · doi:10.48550/arxiv.2010.06266
openalex publication_date 2020/10/13 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
In this paper we investigate the use of model-based reinforcement learning to\nassist people with Type 1 Diabetes with insulin dose decisions. The proposed\narchitecture consists of multiple Echo State Networks to predict blood glucose\nlevels combined with Model Predictive Controller for planning. Echo State\nNetwork is a version of recurrent neural networks which allows us to learn long\nterm dependencies in the input of time series data in an online manner.\nAdditionally, we address the quantification of uncertainty for a more robust\ncontrol. Here, we used ensembles of Echo State Networks to capture model\n(epistemic) uncertainty. We evaluated the approach with the FDA-approved\nUVa/Padova Type 1 Diabetes simulator and compared the results against baseline\nalgorithms such as Basal-Bolus controller and Deep Q-learning. The results\nsuggest that the model-based reinforcement learning algorithm can perform\nequally or better than the baseline algorithms for the majority of virtual Type\n1 Diabetes person profiles tested.\n