Deep Reinforcement Learning based Insulin Controller for Effective Type-1 Diabetic Care
summary
Proposes a DDPG actor-critic insulin controller implemented in Python with SimGlucose + OpenAI Gym, trained and tested against a Padova-model baseline on 30 virtual patients (10 adults, 10 adolescents, 10 children). Runs three simulated scenarios: single meal, clustered multiple meals, and spread-out multiple meals. DDPG uses continuous actions and learns a policy that injects adaptive basal insulin; it consistently reacts faster and keeps glucose nearer target under rapid meal disturbances, while integration of bolus dosing for spread-out meals remains a work in progress. model-free continuous control
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:33Diabetes is on the Rise...
- 2:29Continuous Glucose Monitoring (CGM) and Insulin Pump for Diabetic Control
- 3:52Maintaining Appropriate Blood Glucose Level
- 5:17Closed Loop Control System for Optimal Insulin
- 6:03Reinforcement Learning for Insulin Control
- 6:58Continuous Control with Deep Deterministic Policy Gradient (DDPG)
- 9:39Key Components of DDPG
- 11:24DDPG Algorithm Implementation
- 12:00Implementation and Environment
- 13:05Evaluation Methodology
- 15:44Experiment II: Multi-Meal Frequent Disturbances
- 16:32Experiment Ill: Multi-Meal Disturbances Spread Out
- 17:28Conclusion
related talks
50mDiabetes: Calculating Insulin by Automated Carb Counting Using AI, ML
01h 17mTraining intelligent game agents using deep reinforcement learning
49mThe Wisdom of the Body Revisited
28mDeep Reinforcement Learning: theory, intuition, code
01h 35mDeep Robotic Learning
42mReinforcement Learning using Generative Models for Continuous State and Action Space Systems
01h 29mBuild Your Own Data Science AI Agents
24mDesigning a Video Game AI with Reinforcement Learning