�°IJʹ���

In healthcare, decisions based on clinical text can directly affect patient outcomes. Large language models (LLMs) offer powerful capabilities for analysing such text, supporting tasks like clinical coding, summarisation, and information extraction from electronic medical records. However, their opaque “black-box” nature, combined with occasionally misleading outputs, limited transparency around training data, and the complexity and sensitivity of medical information, raises serious concerns about bias, accountability, and safety. These issues must be addressed for clinical adoption.

This PhD project will investigate methods to make LLMs more explainable and trustworthy in healthcare contexts. It will explore post-hoc interpretability techniques, such as attention visualisation and example-based explanations, as well as approaches for tracing the origin of model outputs.

Candidate Requirements

The ideal candidate will have a strong background in machine learning and natural language processing, with experience in Python and deep learning frameworks (e.g. PyTorch or TensorFlow). Familiarity with large language models, healthcare data, and explainable AI is important for the successful completion of the project. A demonstrated interest in ethical AI and the ability to collaborate in interdisciplinary settings, including with clinical experts, will be highly valued.