My research is focussed on natural language processing (NLP) and interactive machine learning. I am particularly under interested in how to learn models for processing language when faced with uncertainty of various kinds: missing information, noisy labels, changing preferences, weak models, and ambiguous language all lead to uncertainty in how to answer a question when processing some data. 

My current interests include:

• Developing methods for checking the factuality of generated text -- for example, detecting and reducing hallucinations in LLM-generated summaries.
• NLP for healthcare: using AI to help clinicians, patients and researchers access information by analysing clinical notes, personal data and scientific literature.
• Modelling uncertainty in LLM-generated answers: how confident should we be in the model's answers? Using techniques such as Bayesian deep learning to improve uncertainty quantification.
• Learning from limited data: small numbers of preferences, noisy crowdsourced annotations, and synthetic data.
• Adapting small models to specialised tasks such as summarisation and classification, leveraging strong models such as LLMs for distillation.
• Understanding how LLMs represent figurative language and factual information.

See more on our research group page. 

I'm also part of the management team for the Centre for Doctoral Training in Practice-oriented AI. 

Keywords

Natural language processing, NLP, machine learning, summarisation, AI for healthcare, Bayesian methods, interactive AI, interactive learning, crowdsourcing, annotation, preference learning.