Explaining Survival Predictions with Shapley Values

Machine learning models are increasingly used to predict time-to-event outcomes — for example, how long a patient might survive after treatment or when a machine is likely to fail. Unlike standard predictions, these models produce survival curves, which change over time. Existing explanation tools like Shapley values can tell us which features matter, but they … Read more

25-26 On offer: Lip Reading App

This project aims to investigate the relatively new field of lip reading apps (e.g. https://www.youtube.com/watch?app=desktop&v=1xRAG3fHflQ). First thing would be to investigate the field before deciding how to proceed with the development…

Machine learning research topics on big and complex data

I am happy to supervise final-year undergraduate projects and master’s theses (Data Science). Below are the research topics. As these topics rely on advanced mathematical and statistical methods, only candidates with a strong background in mathematics or statistics will be considered. I also welcome PhD enquiries. If you’re interested, please contact me to discuss your … Read more

Evaluating Prototype Explanations in Machine Learning

Prototype-based post-hoc explanations aim to make model predictions interpretable by presenting representative examples (prototypes) that illustrate how the model arrives at decisions. Their evaluation often relies on quantitative metrics such as fidelity (how closely prototypes approximate the model’s decision function), coverage (how much of the input space they represent), stability (whether explanations remain consistent under small perturbations), and diversity (ensuring prototypes capture … Read more

Fairness and Robustness in Risk Detection Models

Risk detection models (such as IBM’s Granite Guardian) are increasingly used to flag harmful prompts and responses in large language model pipelines. These systems are trained on human and synthetic data to identify risks across multiple dimensions, but their reliability and fairness are not guaranteed. They may over-flag certain groups, miss subtle harms, or be … Read more

Debugging Classifications with Counterfactual Explanations

This project investigates how post-hoc counterfactual explanations can be used to debug opaque models such as deep neural networks by revealing which feature changes most influence predictions. In applications like anomaly detection, counterfactuals help clarify why certain cases are flagged as abnormal and expose when models rely on spurious correlations or biased patterns. By using … Read more

Machine learning for fibre disturbance detection via state of polarisation sensing

Project Description: Optical fibre networks are critical to global connectivity, yet they are susceptible to accidental damage or deliberate tampering. While distributed acoustic sensing (DAS) has been widely investigated for disturbance detection, an alternative approach relies on monitoring the state of polarisation (SOP) of light propagating in the fibre. The SOP is highly sensitive to … Read more

Machine learning application to fibre sensing in maritime applications

Project Description: Recent advances in distributed fibre sensing have shown that existing optical fibre cables, originally deployed for telecommunications, can also serve as dense sensor arrays. By sending probe signals through the fibre and analysing the backscattered light, it is possible to extract information about environmental changes such as vibrations, temperature, and strain along the … Read more

4 – Applying attention mechanisms to reinforcement learning in traffic

In reinforcement learning, agents usually process all inputs equally, even if some are more relevant than others. This project will explore the use of a simplified attention mechanism in a car-following environment, allowing a vehicle to focus more on important neighbours (such as the vehicle directly in front). Students will compare the performance of Multi-Agent … Read more

3 – Tackling the cold-start problem in multi-agent reinforcement learning for cooperative driving

Reinforcement learning agents often behave conservatively or inconsistently at the start of training, leading to the so-called cold-start problem. This project will simulate a simple traffic scenario and train a multi-agent RL-based autonomous vehicle controller. Students will then explore strategies to reduce cold-start effects, such as imitation learning for initialization or reusing past experiences, and … Read more