Post-Doctoral Research Associates (PDRAs)

Full Time: 35 hours per week

Fixed term: 14 Months 

The School of Informatics at the University of Edinburgh invites applications for several Post-Doctoral Research Associates (PDRAs), to do research on novel architectures for Large Language Models, under the supervision of Professors Frank Keller, Mirella Lapata, Amos Storkey, and Ivan Titov.

The Opportunity

Five positions are part of the Science of Fundamental AI Research (SOFAIR) Lab, a partnership between UCL and the Universities of Cambridge, Edinburgh, and Oxford. SOFAIR is one of two new national AI research labs funded as part of a £60 million investment from UK Research and Innovation (UKRI). SOFAIR will bring together researchers from across computer science, mathematics, statistics and neuroscience to explore new AI architectures and those designed to run on widely available hardware. This will mean cutting-edge AI will be more widely accessible for everyone, including researchers and institutions.

The PDRAs will be part of the School of Informatics, University of Edinburgh, which is ranked among the top schools in Europe for AI research according to CSRankings. They will be supervised by Professors Frank Keller, Mirella Lapata, Amos Storke, and Ivan Titov, who are leaders in NLP, machine learning, and cognitive modeling. The job holders will collaborate closely with the SOFAIR partners at UCL, Cambridge, and Oxford.

The PDRAs will be responsible for developing new fundamental AI modelling and learning methods that go beyond standard transformers and next-token prediction. The work will explore modular, compositional and compute-adaptive models, including architectures and learning paradigms inspired by neurobiology. The work will also investigate sparse or structured reasoning and new training approaches, such as reinforcement-learning hybrids and gradient-free methods. A central goal of SOFAIR is to improve reasoning, efficiency and interpretability while enabling training and inference across heterogeneous, distributed, small-memory hardware. Specific duties will include designing and implementing novel model architectures, developing scalable training algorithms, and conducting controlled experiments to evaluate reasoning, efficiency, interpretability, and generalisation. The post holders will train and benchmark large models using multi-GPU and distributed-computing infrastructure, including the IsambardAI supercomputer, and will be expected to develop robust research code, analyse model behaviour, and disseminate findings through publications and presentations.

These positions include funding for international travel to attend conferences as well as a dedicated compute allocation on IsambardAI, the UK’s national AI supercomputer comprising over 5000 Nvidia GH200 GPUs. The position is open to UK and international applicants, with visa sponsorship available. This post is advertised as full-time (35 hours per week); however, we are open to considering part-time or flexible working patterns. We are also open to considering requests for hybrid working (on a non-contractual basis) that combines a mix of remote and regular on-campus working.

Advert information