The University of Glasgow is offering a PhD position in Mathematics focused on dissecting deep neural networks, with applications in computational mathematics, computer vision, data analysis, and machine learning. This project aims to advance our understanding of deep learning by exploring the generalization performance of neural networks, the roles of individual layers and substructures, and the effectiveness of optimizers such as sharpness-aware minimization. The research will address the challenges posed by the high-dimensional and non-convex optimization landscapes of neural networks, seeking to enhance the efficiency, explainability, and robustness of these models.

Potential application areas include computer vision and natural language processing, with opportunities to contribute to ongoing AI research collaborations in ecology and climate change. The successful candidate will work under the supervision of Dr Tiffany Vlaar (main supervisor) and Dr Linus Ericsson, both of whom have extensive experience publishing in top venues such as NeurIPS, CVPR, and ICML. The supervisor team offers a strong network and support for publishing research outcomes in leading conferences.

Applicants should possess a degree or higher qualification in a relevant field with a strong mathematical component, such as Physics, Computing Science, Machine Learning, Mathematics/Statistics, or related disciplines. Prior programming experience is desirable, and an interest in machine learning is expected. However, prior knowledge of machine learning is not mandatory, as the necessary background will be provided during the project.

The application deadline is January 2, 2026. Interested candidates should prepare their application materials, emphasizing their mathematical background and programming skills, and submit their application through the University of Glasgow's PhD application portal. For further information, applicants may contact the supervisor team.