PhD Studentship: Low-cost Multi-source Representation Learning

Application deadline: 15/08/2025
Research theme: Computer Science
No. of positions: 1
Eligible for: UK

This 4-year PhD project will be funded by DLA studentship and is open to UK students. The successful candidate will receive an annual tax-free stipend set at the UKRI rate (£20,780 for 2025/26) and tuition fees will be paid. We expect the stipend to increase each year. The start date is 1st October 2025.

Encouraged by the continuing success of modern machine learning (ML) techniques, researchers have become ambitious to develop ML solutions for challenging science and engineering problems with complex input.

For instance, in physics-informed ML, in addition to data examples used by a standard ML setup, domain knowledge serves as an additional input. It can be in an explicit form of rigorous physical laws, or an implicit form of extra data examples collected from physical simulations or their ML surrogates. In medical domains, patient data is typically distributed across multiple hospitals, multi-source learning is used to integrate diverse patient populations to build robust models, but having to protect sensitive information.

Various modern ML paradigms are proposed to address the diverse input needs, accompanied by a boost in algorithmic development, e.g., multi-modal learning, transfer learning, federate learning, and knowledge embedding, etc. However, a significant motivation of applying ML techniques in science and engineering is to accelerate knowledge discovery. It is very much not convenient and time consuming for a user to dive into the ML ocean, searching new techniques to accommodate their own particular input setup and deciding the best modelling practice.

This PhD project will aim at automatic solution development, supporting flexible input setups and addressing in one modelling framework multiple ML tasks as mentioned above, to ease the development burden from users. It will research unified and modular modelling strategies, capable of optimally fusing and aligning diverse types of relevant information sources in representation spaces at both data and model levels, preferably with theoretical guarantees. It will consider variability in data quality, heterogeneity in data sources, information (mis)alignment, domain shift, missing data modality, data privacy, data and computing cost. It will focus on targeted scientific problems to test the solutions, aiming at a lowest development cost and highest solution quality.

Applicants should have, or expect to achieve, at least a 2.1 honours degree or a master’s (or international equivalent) in a relevant science or engineering related discipline. Strong background/skills on machine learning, mathematics, probabilistic modelling and optimisation are preferred.

To apply please contact the supervisor, Dr Mu - Tingting.Mu@manchester.ac.uk. Please include details of your current level of study, academic background and any relevant experience and include a paragraph about your motivation to study this PhD project.

Advert information