
Go to:

David Molyneux
Contact: d.molyneux-2021@hull.ac.uk
PhD topic: Assuring Large Language Models through Knowledge Graph Infusion
David’s research aims to develop methods to assure the trustworthiness and factual accuracy of Large Language Models (LLMs) by infusing them with knowledge graphs. Funded by an EPSRC iCASE Studentship in collaboration with Naimuri – a subsidiary of QinetiQ—and the Engineering and Physical Research Council (EPSRC), this cutting-edge approach will combine neuro-symbolic systems like knowledge graphs (KG) to improve LLM performance.
The research will consider three foundational stages:
-
Developing KG-Based Metrics: Assurance through creation of metrics to detect hallucinations, falsehoods, contradictions, and missing knowledge in LLMs.
-
Enhancing Prompt Engineering: Developing KG-based methods for better prompt engineering, moving us from post hoc assurance to pre-inference assurance.
-
Shallow Infusion of KGs into Training Data: Improving LLM training data with KGs to both enhance performance and move assurance into the pre-training stage.
Additional exploratory stages will consider the impact of infusing KGs into attention mechanisms to improve resilience to counterfactual predictions at different levels of abstraction and independently of training.
The outcome of this research will contribute to building an AI Assurance Framework, offer methods to improve LLMs that fail assurance gates, and provide a basis for the ongoing operational assurance of LLM-based applications.
David has a robust background in data management, assurance and analytics, bringing over a decade of professional experience to his research in AI. He holds an MSc in Artificial Intelligence and Data Science from the University of Hull and a BSc in Mathematics from The University of Manchester.
Before commencing his PhD studies, David worked extensively in data management roles, specialising in master data, data migration, data process development and improvement, data analytics and visualization, data quality and governance, and master data shared services. He has significant experience leading business teams, third-party developers, and shared service teams through ETL, data cleansing, and migration activities. Prior to specialising in data management, David was a qualified internal auditor, focussing on risk management, loss prevention, fraud investigation and auditing and assuring operational systems and processes. His professional experience spans multiple sectors, including consumer goods, transportation, and hospitality/leisure.
Useful Links
RESEARCH INTERESTS
Data Analytics and Visualization
Process Development and Improvement