Towards Reliable AI under Domain Shift: Uncertainty-Aware Learning

Applications are invited for a self-funded, 3 year full-time or 6 year part-time PhD project. 

The PhD will be based in the School of Computing, Mathematics and Physics and will be supervised by Dr Dongxu Gao. 

Modern AI systems in computer vision achieve strong performance under controlled conditions but often fail to generalise in real-world environments due to domain shift, where data distributions differ across settings. This limits their reliability and safe deployment.

 This PhD project aims to develop reliable AI models for general imaging tasks that remain robust under such distributional changes. The research will focus on integrating uncertainty-aware learning, domain generalisation, and adaptive inference to enable models to recognise their limitations and respond effectively to new environments.  The project will initially investigate these challenges in general computer vision tasks (e.g., recognition, segmentation), with methodologies designed to be broadly applicable. As the research progresses, the approaches may be extended to high-impact domains such as medical imaging, where robustness and trustworthiness are essential. 

Artificial intelligence (AI) models for image analysis have achieved strong performance in tasks such as classification and segmentation. However, these models often fail to generalise in real-world environments due to domain shift, where data distributions differ across settings (e.g., changes in sensors, environments, or acquisition conditions). This limitation poses a major challenge for the reliable deployment of AI systems.

This PhD project aims to develop reliable and uncertainty-aware AI models for general imaging tasks that can operate robustly under such distributional changes. The research will focus on integrating uncertainty estimation, domain generalisation, and adaptive inference strategies, enabling models to quantify their confidence, recognise potential failure cases, and respond effectively to new environments.

The project will initially investigate these challenges in general computer vision problems, such as object recognition and segmentation across multiple domains, to develop broadly applicable methodologies. It will explore approaches that go beyond standard supervised learning, including techniques for improving generalisation from limited data and lightweight adaptation at test time.

The outcomes are expected to contribute to the development of trustworthy and deployable AI systems, with potential extension to high-impact domains such as medical imaging.

Fees and funding

Visit the research subject area page for fees and funding information for this project.

Funding Availability: Self-funded PhD students only.

PhD full-time and part-time courses are eligible for the UK Government Doctoral Loan (UK and EU students only - eligibility criteria apply).

Bench fees

Some PhD projects may include additional fees – known as bench fees – for equipment and other consumables, and these will be added to your standard tuition fee. Speak to the supervisory team during your interview about any additional fees you may have to pay. Please note, bench fees are not eligible for discounts and are non-refundable. 

How to apply

We’d encourage you to contact Dr Dongxu Gao ([email protected]) to discuss your interest before you apply, quoting the project code.

When you are ready to apply, please follow the 'Apply now' link on the Computing PhD subject area page and select the link for the relevant intake.. Make sure you submit a personal statement, proof of your degrees and grades, details of two referees, proof of your English language proficiency and an up-to-date CV.  Our ‘How to Apply’ page offers further guidance on the PhD application process.
 

When applying please quote project code CMP10570529.