Towards Next-Generation Healthcare Ecosystems: Safe Intelligence Augmentation via Machine Learning and Contrained Control

In this talk, he will explain how to synthesize machine learners and constrained controllers to provide solutions that are capable of handling one or more combinations of safety, resource, and compatibility constraints. Concretely, he will discuss solutions for: (1) deploying model predictive control formalisms on resource-limited wearable artificial pancreas systems; (2) enabling clinical decision support mechanisms with deep neural networks in the control loop; and, (3) generating cautious data-driven control policies via safe reinforcement learning. He posit that these solutions are key steps towards safe intelligence augmentation for decision-making in next-generation healthcare ecosystems.