Join a focused part of a current project or propose a related idea. Agree the scope around your interests, preparation, degree requirements, and available resources.
Support acknowledgements below describe project support; they do not imply that a funded student position or scholarship is available. Discuss current opportunities with the supervisor.
AI and Digital Public Infrastructure (DPI)
Supported by the Gates Foundation.
We explore how AI can support trustworthy and useful digital public infrastructure.
- Facial verification and liveness detection: Verify identity and distinguish a live person from a presentation attack, with careful evaluation of errors and robustness.
- AI agents for DPI: Investigate agents that help people navigate DPI services and workflows, with appropriate permissions, verification, and human oversight.
- Verifiable credential (VC) use cases: Explore how verifiable credentials can support practical service interactions and how their use should be evaluated.
Possible starting point: Agree on one verification task, agent workflow, or credential use case; define the baseline and evaluation criteria before building.
Traffic Crash Analysis
Supported by Toyota Mobility Foundation.
We combine computer vision with traffic-conflict and crash-analysis concepts to understand risky road interactions. A particular interest is time to collision (TTC): an estimate based on the motion assumptions used, rather than a direct measure of crash probability.
- Detect and track road users in video.
- Estimate trajectories and measures such as TTC under clearly stated assumptions.
- Analyse conflicts and crash sequences, including how tracking errors and scene conditions affect the conclusions.
Possible starting point: Evaluate one tracking or conflict-measurement component on an agreed video dataset, including its failure cases.
Sign Language
Supported by NBTC.
We develop deep-learning approaches for sign-to-text and text-to-sign. These tasks require attention to language structure, temporal information, and the usefulness of the output to sign-language users.
- Build and evaluate models for recognising or translating signed sequences into text.
- Investigate representations and generation methods for producing signed output from text.
- Evaluate meaning, intelligibility, and limitations with suitable data and, where feasible, sign-language users.
Possible starting point: Reproduce an agreed baseline and analyse its errors on one defined task. The target sign language and available data will be agreed with the team.
Honest and Adaptive Sales Agents
Supported by insurance companies.
We investigate how to build sales agents that are honest, ethical, effective, adaptive, and capable. Effectiveness includes helping customers make informed decisions and finding a suitable outcome.
- Ground product explanations in verified information and communicate uncertainty and limitations.
- Adapt explanations and questions to customer needs without deception, pressure, or exploitation of vulnerability.
- Evaluate task completion, factual accuracy, suitability, transparency, and respect for customer choice.
Possible starting point: Define a bounded sales conversation and build an evaluation set that tests both useful assistance and behaviour under pressure to mislead.
Finding Your Role
Master’s students will usually own a focused component within a PhD or grant project. PhD students develop a broader original contribution, which may connect with one of these efforts. Specific roles, data access, resources, supervision, and any funding arrangements are agreed individually.
Read Getting Started to prepare a one-page idea or project-interest note. See Research Areas for the lab’s broader directions and degree expectations.