Back
NeuroSys Projekt D (Teilprojekt D4) „Impulses through applications - Individualized medical controllers“
The aim is to develop an invasive reinforcement learning controller for optimized and closed-loop treatment of Type 1 Diabetes. The future goal is to have an artificial pancreas that runs eventually on neuromorphic hardware. The Problem is modelled as a Markov decision process and neural networks models are trained in simulation with a virtual patients model, getting the blood glucose measurements and deciding the insulin amount. The project contributes to exploring both the performance benefits and limitations of neuromorphic hardware when used in practical medical devices. The focus is on responsible AI integration throughout the research pipeline and the aim is to create efficient, private, and secure methods transferable across applications. Data centric green AI approaches are investigated. An ECG classifier is used to integrate energy efficiency in machine learning methods while maintaining data privacy. A modified split-learning mechanism and quantization methods are used for this purpose.
Field of action:
Neuromorphic Computing > Algorithms > Machine Learning
- Faculty 6 – Electrical Engineering and Information Technology
- Profilbereich Information and Communication Technology (ICT)
Address:
Lehrstuhl für Informationstheorie und Datenanalytik, Kopernikusstraße 16, 52074 Aachen
Contact:
Prof. Anke Schmeink
Anke.schmeink@inda.rwth-aachen.de
Homepage:
https://www.inda.rwth-aachen.de/
Status:
running