Postdoctoral Research Associate in the Mathematical and Computational Foundations of Artificial Intelligence

We invite applications for a Postdoctoral Research Associate (PDRA) to join the EPSRC Hub on the Mathematical and Computational Foundations of Artificial Intelligence. Applicants are expected to have published in leading machine learning conferences or similar venues. One or two PDRAs will be recruited to work within one of, or across, the four research themes: Learning with Structured & Geometric Models, Low Effective-dimensional Learning Models, Implicit Regularization, and Reinforcement Learning through Stochastic Control (a brief description of each these is as follows (additional details are in the further particulars):
Learning with Structured and Geometric Models. We will apply tools from manifold learning and Riemannian optimisation to leverage the underlying manifold structure for better training and novel network designs.
Low Effective-dimensional Learning Models. We will extend foundational theory of how large ML systems can be regularised to have dramatically fewer trainable parameters without sacrificing accuracy by analysing the use of low-dimensional building blocks
Implicit Regularization. We aim to develop mathematical understanding of implicit regularisation properties in deep neural networks to guide the development of algorithmic paradigms aimed at combining statistical optimality with computational efficiency.
Reinforcement Learning through Stochastic Control. We will develop methods from stochastic control, which will provide a mathematically grounded approach that has a well-posed continuous-time limit (as opposed to traditional RL methods that are inherently discrete and do not scale favourably for high frequency observations without judicious hyper-parameter tuning).
There are two two-year, fixed-term positions, funded by a research grant from the EPSRC and the Mathematical Institute, and one will have emphasis on optimization aspects of the above themes. The start date for this post is flexible.
The successful candidate will be expected to conduct research which falls within the remit of this large-scale project and will have the opportunity to do so collaboratively with other members of the hub, both at Oxford and/or with hub partners which include universities as well as companies and governmental organisations.
They will contribute to the activities of the wider machine learning and data science research group and write up the results of their work, with co-authors, for publication in refereed journals and proceedings. There will be opportunities to contribute a small amount of teaching to the department, of at most three hours a week during the academic terms.
We proudly hold a departmental Athena SWAN Silver Award and an institutional Race Equality Charter Bronze Award.

Postdoctoral Research Associate in the Mathematical and Computational Foundations of Artificial Intelligence
Andrew Wiles Building, Radcliffe Observatory Quarter, Woodstock Road, Oxford, OX2 6GG.