PhD Studentship: Reliable and Generalisable Knowledge Editing in Large Language Models

Overview

This PhD studentship focuses on reliable and generalisable knowledge editing in large language models within the School of Computing Sciences at the University of East Anglia. The project is supervised by Dr Farhana Ferdousi Liza and aims to develop and evaluate methods for reliable, localised, and generalisable knowledge updating in large language models. Research activities will involve adapting established editing methods such as ROME and MEMIT, comparing them with retrieval-augmented generation, external memory mechanisms, and conventional fine-tuning, and assessing performance across benchmarks like CounterFact and zsRE. Students may also investigate knowledge provenance and deploy selected models in embodied AI systems or social robots to conduct human-AI interaction experiments.

Who can apply

Funding for this studentship is available to UK applicants as part of a competition for a Faculty of Science funded studentship. Applicants must meet the standard minimum entry requirement of a 2:1 degree in Computer Science.

What you get

The funding provides home tuition fees and an annual stipend for three years. The current institutional PhD stipend rate is £21,805 per annum, which is subject to any agreed inflationary uplift for the 2027 to 2028 academic year.

Requirements

Applicants are required to hold a 2:1 in Computer Science as the standard minimum entry requirement. The mode of study is full-time, and the work arrangement is on-site in the United Kingdom.

Key dates

The closing date for applications is 2027-02-24. The start date for the program is 1 October 2027.