featured

We introduce GlyTwin as a novel digital twin framework designed to support individuals with Type 1 Diabetes through personalized and actionable behavioral interventions. Rather than only predicting glucose outcomes, GlyTwin generates counterfactual treatment recommendations that suggest minimal changes in insulin dosing, carbohydrate intake, and insulin timing to help prevent hyperglycemia.

We present a contextual bandit–based reinforcement learning framework for data-driven tuning of Model Predictive Control (MPC) for personalized physical activity interventions. A TD3-based agent learns MPC tuning parameters that improve step-goal tracking and self-efficacy while preserving the feasibility and constraint-handling capabilities of MPC.

This work focuses on improving blood glucose prediction by incorporating glycemic-aware training strategies that better capture hypo- and hyperglycemic events.