RiTuaL: Reliable and Trustworthy Machine Learning for Time Series Modelling in Mobile Health

Mobile health (mHealth) applications are reshaping the future of healthcare by seamlessly monitoring environmental factors (like light intensity) and personal attributes (such as heart rate, motion, and location). These apps provide real-time health predictions and analytics, enabling long-term, out-of-clinic monitoring without requiring active participation from users. These applications need to be safe in terms of: 

i) Ensuring health predictions remain precise across various contexts. 

ii) Providing reliable output confidence for each prediction. 

Despite the widespread use of machine learning in mHealth, maintaining safety and accuracy in dynamic, real-world conditions remains a significant challenge. This is due to natural shifts in data distribution, which are particularly complex in sequential mHealth data: 

i) Covariate shifts: Variations in sensor placement, environment, sensor calibration, and demographic differences. 

ii) Concept drifts: Changes due to disease progression and evolving health conditions. 

 In this exciting project, you will work on enhancing the accuracy and reliability of mHealth predictions with the long-term goal of translating your findings into clinical practice. You will tackle key research questions, such as: 

i) Quantifying Data Distribution Shifts: 

– How can we quantify shifts in data distribution for mHealth sensing modalities? 

– Develop methods to connect empirical advancements with theoretical understanding. 

ii) Building Robust mHealth Models: 

– How can we create mHealth models that remain accurate despite distribution shifts? 

– Explore representation learning and domain adaptation to enhance model robustness. 

iii) Reliable Uncertainty Estimates: 

– How can we ensure mHealth models produce reliable uncertainty estimates even with data shifts? 

– Integrate advanced uncertainty quantification methods to improve model reliability. 

This project brings together expertise in time series modeling, mobile sensing, and machine learning reliability to address critical challenges in mHealth. By joining our team, you will have the opportunity to work on pioneering research that has the potential to make a significant impact on healthcare. 

Supervisors: Dr Ting Dang (University of Melbourne) and Dr Abhirup Ghosh (University of Birmingham)

This Joint PhD project will be primarily based at the University of Melbourne with a minimum 12-month stay at the University of Birmingham.

Eligibility Criteria

The entry requirements for the Birmingham/Melbourne Joint PhD are either:

An upper second-class four-year honours UK undergraduate degree in a relevant subject (or equivalent)

An MSc/MRes in a relevant subject

How to Apply

Applicants must make their application through the application portal. Applicants are encouraged to contact prospective supervisors informally to discuss the project. Please detail the supervisor and project title under the Research Information section of the application form.

Please note that applications will close once a suitable candidate is identified. Interested applicants are encouraged to submit their application as early as possible.