PhD Studentship: Efficient Deep Surrogate Models for Inverse Problems

Project title: Efficient Deep Surrogate Models for Inverse Problems

Project contact: Dr Taysir Dyhoum

Funding info:

Home fees (2023/24) included plus an annual stipend paid at the UKRI rate (£17,668 for 2022/23). 

Mode of study: Full time

Eligibility: Open to home & overseas students.

Eligible overseas students will need to make up the difference in tuition fees. 

Key dates:        

Closing date: 16 October 2023

Expected start: January 2024

Project summary

Applying imaging techniques for industrial or biomedical applications frequently requires solving non-trivial inverse problems that need to be solved numerically. Such inverse problems are sought to reveal object properties which cannot be observed directly from the measurements. The inverse problems can be seen as an inversion of the forward or observation model. By the nature of the corresponding experimental setup, the inverse problems are ill-posed or severely ill-defined (i.e., The solution may not exist, or if exists it may be very sensitive to the observation errors). Even when an ill-posed inverse problem can be re-defined through regularisation it still poses challenges as resulting regularisation parameters need to be estimated to find a compromise between the robustness & fidelity of the solution.

Therefore efficient, accurate, stable & reliable approaches are required but these are difficult to obtain analytically. The added difficulty, after discretisation of the problem, is typically a very high dimensionality of the non-observable space, e.g., in medical image reconstruction, the dimensionality of the “hidden” space can easily exceed 1M. Hence the proposed project will integrate machine/deep learning into the existing mathematical/statistical approaches to speed up knowledge-based modelling – i.e. PDEs solvers.

Aims and objectives

The project will be set up in a generic context of solving inverse problems, with methods validation using simulated as well as established real MRI & CT, and Electrical Impedance Tomography EIT problems.

Specific requirements of the project

Essential:

• Minimum bachelor’s degree (or equivalent qualification) at 2:1 level or above in mathematics, physics, computer sciences, biomedical engineering, or similar science area. EU/International applicants require an English Language level of UKVI IELTS6.5 (no sub-score less than 6.0).
• Prior knowledge of Numerical Analysis, Partial Differential Equations, and Bayesian statistics or Process-based Modelling.
• Good analytical skills & experience programming with multiple languages (C++, MATLAB, Python, Java, etc)
• Excellent written and oral communication skills and ability to keep accurate records of research results & activity.
• Effective time management and prioritisation of work. Commitment to meeting deadlines. Willingness to undertake any necessary training for the study.
• Self-motivated ability to work independently and as a part of a team.

Desirable:

• UK Master’s Degree (or equivalent qualification) in a relevant area.
• Working experience in artificial intelligence, machine learning, computer vision or image processing.
• Coding skills
• Evidence of independent research work.

How to apply

Interested applicants should contact Dr Taysir Dyhoum for an informal discussion.

To apply you will need to complete the online application form for a full-time PhD in Computing and digital technology (or download the PGR application form), by clicking the 'Apply' button, above.

You should also complete the PGR thesis proposal (supplementary information) form addressing the project’s aims and objectives, demonstrating how the skills you have maps to the area of research and why you see this area as being of importance and interest. 

If applying online, you will need to upload your statement in the supporting documents section, or email the application form and statement to mailto:PGRAdmissions@mmu.ac.uk.

Please quote the reference: SciEng-TD-2023-deep-surrogate

Advert information