Data analytics and machine learning

In today’s data-driven world, the abundance of information from various sources, such as databases, streams, and Internet of Things (IoT) devices, demands efficient methods for analysing, organising, sharing, and displaying it.

Our group’s data analytics division tackles this challenge through a combination of ensemble and federated learning methods, approximation techniques, optimised allocation of tasks within distributed computing environments, network analysis and applied data science.

Our machine learning team focuses on developing cutting-edge algorithms capable of predicting and interpreting diverse datasets, spanning from image, Natural Language Processing (NLP), and tabular data.

Aims

Our objective is to pioneer cutting-edge learning and reasoning algorithms, aiming to emulate brain-like intelligence within resource-constrained environments.

We prioritise responsible AI practices and seek to cultivate interdisciplinary partnerships with experts spanning security, healthcare, finance, transport, socio-economics, and environmental science, driving solutions to real-world challenges, and advancing interdisciplinary research endeavours.

We contribute to teaching by exposing students to state-of-the-art practices in data analytics and machine learning.

Research

There are two branches to our group.

Data Analytics

Our work in data analytics largely focuses on efficient large-scale implementation of machine learning models across distributed environments, and effective deployment of machine learning systems across various platforms:

Partitioning machine learning algorithms across distributed environments (such as cloud and IoT) for both real-time (stream processing) and batch learning.
Ensemble and federated learning across distributed systems platforms.
Container frameworks (such as Docker, Kubernetes) and in-network capability (such as Network Function Virtualisation and MiddleBox approaches) to support machine learning.
Approximation techniques to trade off accuracy and execution time (convergence time).
Network analysis.

Machine Learning

Our expertise in machine learning involves fundamental research on new algorithm development and applied research on solving real-world problems to benefit the economy and society:

Robustness: Building models that are resilient to errors, noise, and unexpected data distributions.
Lifelong and continual learning: Investigating methods for continual learning and lifelong adaptation, enabling models to learn from new data and tasks over time.
Explainable AI: Creating models that allow us to understand their decision-making processes, build interpretable models of the real world, and foster trust in their applications.
Limited data: Leveraging limited data effectively and building generalisable models.
Scalable and resource-efficient methods: Exploring techniques for handling massive datasets effectively that operate in real-time with limited computational resources.
Reinforcement learning and control: Developing robust learning algorithms that enable intelligent agents to make optimal decisions and effectively interact with their environment.
Ethical and societal implications: Investigating the ethical implications of machine learning algorithms and developing frameworks for responsible AI deployment, ensuring fairness, transparency, and accountability.
Development of machine learning toolboxes.

Applications

Disaster Management: Producing models to identify critical infrastructure components, guiding protection strategies against disasters. For example, in underdeveloped areas post-disaster, a last-mile distribution operations model has been designed to ensure aid reaches affected communities efficiently. A model that defines mitigation operations for transport infrastructure has been proposed, reducing disaster impact. Moreover, a model that plans fuel management strategies has been devised to curb wildfires, safeguarding vulnerable regions.
Policing and Public Security: Developing fair patrolling strategies, considering population distribution, and fostering equitable law enforcement. A tool has been created to detect and visualise hate speech on social networks, aiding in combating online toxicity. Additionally, responses to family/gender violence have been analysed, informing intervention strategies. Models have also been formulated to evaluate gender violence recidivism, enhancing preventive measures.
Healthcare: Using health data to detect and treat diseases effectively, improving patient outcomes through precision medicine approaches.
Financial Technology: Developing innovative data models to solve problems in financial services including capital markets, insurance, and customs.
Quantum Technologies: Robust and interpretable reinforcement learning can aid in developing high-fidelity and robust quantum controls, crucial for practical application of quantum devices.
Medical Diagnostics: Novel magnetic resonance imaging and spectroscopy techniques facilitated by control and reinforcement learning can lead to improved medical diagnosis and a deeper understanding of the body’s biochemical processes through machine learning.
Transport: Simulating new modes of transport and new challenges in reaching Net Zero carbon emissions

Projects

AI Hub in Generative Models (Funder: EPSRC)
Enhancing Insurance Brokerage with Advanced Data Solutions, (Funder: Innovate UK)
Hartree Centre Cardiff Hub (Funder: STFC)
Optimal Lead Allocation for Insurance Services (Funder: Innovate UK)
The High Streets Task Force (Funder: DLUHC)
Road Traffic Scenarios (Funder: Foundation for Integrated Transport via Green Alliance)
Accelerating Modal Shift (Funder: Foundation for Integrated Transport via Green Alliance)
Spatial Design Network Analysis (SDNA) software (various funders)