Project: Reinforcement Learning (RL) is a Machine Learning (ML) paradigm, used for decision-making tasks, where an agent learns to achieve a specified goal, by interacting with its underlying environment. Research related to RL-based glucose control systems is relatively minimal compared to popular RL tasks (games: Chess, Go, physics simulations: MuJuCo). The development of RL-based glucose control systems requires ground up development resulting in significant experimentation overhead. This project aim is to facilitate the development of RL-based control algorithms by providing a high-performance simulation environment for experimentation by following an end-to-end GPU-based implementation using the PyTorch framework. The RL algorithms include both Deep Learning (DL) and more traditional ML approaches.
Timeline:
- Background & Related Work. Learn about Type 1 Diabetes simulators and familiarize yourself with our existing codebase (https://github.com/chirathyh/GluCoEnv). Then write a literature survey to explore relevant algorithms and systems based on their applicability to the target problem.
- Coding. Explore opportunities to improve efficiency, develop visualization scripts, and additional functionalities for researchers/developers.
- Experiments & Validation. Benchmark the simulator and assess its performance (parallelization, simulation throughput).
- Bonus Item. Integrate the developed simulator into our demonstration system called CAPSML at https://capsml.com/ for others to try out your work OR explore implementation using JAX/TensorFlow frameworks.
- Conclusion. Thesis writing, potential publication writing, and code documentation.
Supervisors:
- Dr. Chirath Hettiarachchi (Research Fellow, School of Computing), a computer scientist working on ML and Reinforcement Learning algorithms to develop closed-loop treatment for clinical applications.
- Dr. Lex van Loon (Research Fellow, College of Health and Medicine), a Technical Physician working on physics informed ML for the creating for Medical Digital Twins.
- Prof. Hanna Suominen (Associate Director of JEI, Neuroinformatics), a computer scientist with 20 years’ of experience at forefront of bringing ML/DL algorithms, document analysis methods, and personalized medicine technologies to bear.
Requirements: To undertake substantial work on this project, students must have a solid background and very good skills in software development, programming, experience using Python, PyTorch, and ML/DL. To apply for this opportunity please forward your CV and transcript to Dr. Chirath Hettiarachchi (chirath.hettiarachchi@anu.edu.au). Successful candidates will be invited for a discussion to learn more about the project.