Applications accepted all year round. Competition Funded PhD Project (Students Worldwide).
About the Project
Optimization algorithms are fundamental to modern data science and machine learning, underpinning applications from training AI models to solving large-scale inverse problems. However, traditional optimization methods are usually based on fixed update rules and often require substantial manual tuning.
This PhD project will investigate Learning to Optimize (L2O), where machine learning is used to automatically design or improve optimization algorithms. The project will explore how components such as step sizes, momentum, preconditioners, update directions, proximal operators, or optimization geometries can be learned from data.
A particular focus will be on developing structured and interpretable learned optimizers that combine the flexibility of machine learning with the reliability of classical optimization [1,2].
The project will study both the theoretical properties of learned optimization algorithms, such as convergence, stability, and generalization, and their practical performance on applications arising in machine learning, computational imaging, generative modelling, and statistical inference.
The project is suitable for students with a strong background in applied mathematics and statistics (or computer science), and an interest in optimization and machine learning.
If you are interested in this project, please email your CV and transcript to Dr Junqi Tang (j.tang.2@bham.ac.uk).
Funding Notes
For UK and EU candidates:
Funding may be available through a college or EPSRC scholarship in competition with all other PhD applications: https://www.birmingham.ac.uk/research/centres-institutes/research-in-mathematics/mathematics-phd-information
Strong candidates are encouraged to make an informal inquiry.
For non-UK/non-EU candidates:
Strong self-funded applicants will be considered.
For Chinese candidates:
- The China Scholarship Council (CSC) Scholarship: https://www.csc.edu.cn/chuguo
- China Scholarship Council (CSC) PhD Scholarships Programme at the University of Birmingham
- PhD Placements and Supervisor Mobility Grants China-UK: https://www.britishcouncil.cn/en/programmes/education/higher/opportunities/phd
References
- Tan, H.Y., Mukherjee, S., Tang, J. and Schönlieb, C.B., 2023. Data-driven mirror descent with input-convex neural networks. SIAM Journal on Mathematics of Data Science, 5(2), pp.558-587.
- Tan, H.Y., Mukherjee, S., Tang, J. and Schönlieb, C.B., 2024. Boosting data-driven mirror descent with randomization, equivariance, and acceleration. Transactions on Machine Learning Research.
Project Supervisors
Dr. Junqi (Billy) Tang
Career overview: Dr Junqi (Billy) Tang received his MSc and PhD at the University of Edinburgh in 2015 and 2019, respectively. After completing his studies, he joined the University of Cambridge as a Research Associate, focusing on stochastic optimisation and medical imaging. In March 2023, he was appointed as an Assistant Professor in Mathematical Optimisation and Data Science at the University of Birmingham's School of Mathematics. Dr Tang's research interests encompass large-scale optimisation and learning theory, particularly in the theoretical foundations of non-convex optimisation in machine learning, data-driven optimisation, and efficient deep unrolling networks in computational imaging.
Research interests: Dr Junqi Tang's research interests encompass large-scale optimisation and learning theory, particularly in the context of data science. His recent work has concentrated on the theoretical foundations of non-convex optimisation within machine learning, data-driven optimisation, and the development of efficient deep unrolling networks for computational imaging.