PhD Studentship: Understanding Deep Learning

PhD Supervisor: Antonia Marcu

Supervisory Team: Antonia Marcu, Jonathon Hare

Project description:

Deep Learning (DL) is a widely successful tool. However, there are many fundamental challenges left to solve in DL. One such challenge is “simplicity bias’’, which is a model’s bias towards learning simplistic rules even at the cost of performance. As a result, models often rely on spurious correlations and fail to generalise. For example, a model might learn to identify cows by the presence of grass in the background and fail to generalise to other landscapes. More worryingly, a model might provide a medical diagnosis based on a similarly spurious rule such as visible marks left by medical previous treatments. It is therefore difficult to ensure that the models we train are reliable.

This project will research and develop new notions of model quality and tools for measuring them. We will analyse learned representations and explore how their properties relate to generalisation and robustness. Creating new notions of model quality will help further understand DL models, which in turn will lead to the creation of better, more reliable, and more interpretable machines. This additionally has the potential to set the foundations for an empirical theory of generalisation and revolutionise the way we think about model generalisation.

Why join our team?

• Innovation: We are dedicated to doing good science first and foremost. You will conduct rigorous research with a focus on creating new insights.
• Development: We will help you develop a wide range of skills and publish in the top conferences.
• Ownership: Discover and integrate your research vision into the project and make a real impact.
• Community: We value cooperation and foster individual respect. Engage in group activities in a nurturing and intellectually stimulating environment.

Who are we looking for?

• Passionate & Creative: A driven person ready to tackle challenging problems, innovate, and aspire to publish at the top of the field.
• Methodical & Resilient: A strong analytical thinker, resilient in overcoming challenges.
• Open & Curious: An adaptable and communicative person with a desire to continuously learn.

What do we expect?

• Undergraduate-level background in theoretical ML and DL
• Programming experience (ideally Python, PyTorch)
• Experience with ML/DL projects

If you wish to discuss any details of the project informally, please contact Antonia Marcu at a.marcu@soton.ac.uk

Entry Requirements

A very good undergraduate degree (at least a UK 2:1 honours degree, or its international equivalent).

Closing date: 31 August 2024.

Applications will be considered in the order that they are received, the position will be considered filled when a suitable candidate has been identified.

Funding: We offer a range of funding opportunities for both UK and international students, including Bursaries and Scholarships.  For more information please visit PhD Scholarships | Doctoral College | University of Southampton

Funding will be awarded on a rolling basis, so apply early for the best opportunity to be considered.

How To Apply

Apply online via the Apply button. Select programme type (Research), 2024/25, Faculty of Engineering and Physical Sciences, next page select “PhD Choose an item. (Full time)”. In Section 2 of the application form you should insert the name of the supervisor: Antonia Marcu

Applications should include:

• Research Proposal
• Curriculum Vitae
• Two reference letters
• Degree Transcripts/Certificates to date

For further information please contact: feps-pgr-apply@soton.ac.uk

Advert information