With the advancement of deep learning, Explainable AI (XAI) has become a critical research field which aims to deliver human-understandable insights into the working mechanisms of deep learning models. While many XAI techniques have been developed, they cannot be directly applied to the more advanced deep learning models such as large language models, due to their significant complexity. Moreover, most XAI techniques have remained as post-hoc explanations rather than a way to enhance the accuracy of the deep learning models.
In this project, we will conduct research into XAI for large language models. This will involve a comprehensive literature review of current state-of-the-art methods and conduct experimental studies using open-source toolkits and public datasets. The aim is to establish a comprehensive benchmark of state-of-the-art XAI methods with critical analysis of their performance for various applications, such as reasoning and question answering.
School
Computer Science and Engineering
Research Area
Artificial Intelligence | Deep learning | Large language models
- Research environment
- Expected outcomes
- Supervisory team
- Reference material/links
You will work closely with the academic supervisors and other PhD students in the group.
Software development for experimental analysis of XAI methods on large language models; and a report summarising the literature review and experimental results of current state-of-the-art XAI methods.
