Honours and Masters project

Passive acoustic monitoring is a well-established, non-invasive technique for wildlife monitoring, with growing interest in bioacoustic individual-level recognition—the ability to distinguish individual animals based on their vocalisations. While existing approaches perform well when all individuals are known a priori, wildlife populations are inherently dynamic, making such closed-set assumptions unrealistic in natural environments.

This project aims to design and implement an Agentic AI for Human Assistance (the “assistant agent”) to support everyday information work, e.g. email coordination. Unlike conventional chatbots, the proposed system follows an agentic workflow: it can plan, use tools, and execute multi-step tasks, while incorporating Human-in-the-Loop (HITL) mechanism at critical decision points to ensure trustful, safety, controllability, and explainability.

In today's digital landscape, cyberattacks are increasingly impacting organisations by disrupting critical services and compromising sensitive data. As these attacks grow in volume and complexity, security teams are increasingly challenged to safeguard sensitive data and maintain operational continuity. Manual efforts of security assessment often led to inconsistent and delayed results, high operational costs, and increased window of opportunity for potential attackers. To effectively mitigate these risks, there is a pressing need for automated security assessment.

As cities face unprecedented growth, the need for tools that can integrate diverse knowledge sources ranging from geospatial data to the nuanced intuition of urban planners is critical. This research will explore how Bayesian Networks can be adapted to serve as configurable, transparent models that empower decision-makers to weigh alternatives involving complex factors such as development yield, urban zoning, locality to services and infrastructure capacity.

Communication networks show interaction between people over time, and are key to the identification of criminal networks and criminal activity. This project will investigate how future events might be able to be predicted, based on dynamic graphs representing prior interpersonal communications. The project will consider (a) how Graph Neural Networks can best be used for this machine learning task; (b) how visualisation techniques can best depict both known-past and predicted-future events.

Project Description

Teamwork is a big part of university life, but not all teams work smoothly. Students often face issues such as uneven contributions, unclear communication, or members falling behind. Teaching staff receive a large amount of peer feedback. But the information is often dispersed across multiple reports and can be time-consuming to interpret—particularly in large cohorts. A system that could automatically identify which teams are struggling, and why, would allow educators to offer timely, targeted support.

This project examines how films produced in Asian markets perform in terms of commercial success and critical recognition using real-world industry data. Students will compile a dataset of films from regions such as Hong Kong, China, South Korea, and Southeast Asia, drawing on publicly available sources to analyse indicators such as production budget, box office revenue, streaming platform release, and awards. Using quantitative data analysis methods, the project aims to identify patterns and factors associated with successful film outcomes.

Multiple master thesis/honour projects on MLLM/ human understanding are available.

Benefits: we aim for publication at top conferences and journals. You will have chance for full PhD scholarship. For those working hard, paid RA opportunities will also be provided.

requirement: WAM>80 and high self-motivation 

Inclusive learning environments are essential for ensuring that all students can fully participate and succeed. Neurodivergent learners—including those with ADHD, autism, dyslexia, and other cognitive differences—often navigate university settings that were not designed with their learning profiles in mind. Challenges related to feedback interpretation, cognitive load, communication, and assessment design can create barriers that impact learning, confidence, and wellbeing.

Autoregressive moving average (ARMA) models remain a competitive tool for forecasting low signal-to-noise ratio time series, due to their flexibility, low complexity and physical plausibility. They predict the next observation in a time series as a linear combination of a number of previous observations as well as a number of hidden (latent) random innovations. The AutoARIMA package remains a staple benchmarking tool against which forecasting techniques must be compared.