Leading AI innovation from fundamental research to societal impact
The school is home to leading research and teaching within the area of artificial intelligence (AI). We develop state-of-the-art techniques in key AI areas such as machine learning, computer vision, robotics, evolutionary computation, optimisation, multi-agent systems, natural language processing, predictive analysis, virtual reality. We apply those techniques to solve challenging problems in health, agriculture, biology, energy, education, mining, cybersecurity, defence, space, and other areas of high importance to society.
Machine Learning Foundations
Research in machine learning within the School focuses on the theoretical principles that govern how Artificial Intelligence systems and their models learn, generalise, and remain robust in complex environments. Our work includes optimisation algorithms, statistical learning theory, adapting pre-trained models to changing problems, and the behaviour of models in high-dimensional, low-resource, and low-data settings. Our researchers investigate large language models, examining how they represent knowledge, their reliability, and where their limitations arise, and develop techniques for efficient execution in resource-constrained settings. Research in explainable AI addresses the “black box” issue and develops interpretable model behaviour, and research on causal machine learning methods uncovers cause–effect relationships in data, thereby improving explainability and the transferability of pre-trained models to other domains. These combined efforts support applications where understanding is as important as performance. In healthcare, they enable clinically meaningful predictions and treatment insights. In defence and autonomous systems, they improve robustness and build trust in decision-making. In software engineering, they support debugging, validation, and the reliability of AI-driven systems.
AI-based optimisation
Research in optimisation focuses on artificial intelligence and optimisation methods that are frequently used to solve hard and complex optimisation problems. This includes approximation algorithms, dynamic programming, integer linear programming, evolutionary computation, local search, and their combination with machine learning and data analytic techniques. In our research related to real-world applications, we pay particular attention to constraint-handling techniques, multi-objectivity, and dynamic environments. In our theoretical work, we analyse how evolutionary computation and other search methods from the area of artificial intelligence work and show in a rigorous way how they can deal with different types of problems. Such studies are important for the trust in AI-based optimisation methods.
Knowledge discovery and responsible AI
Research in data mining focuses on uncovering structure and meaning in complex datasets, including feature selection, subgroup discovery, and methods for identifying significant patterns in high-dimensional data. This area is closely linked to causal reasoning, ensuring that discovered relationships are meaningful and actionable. Responsible AI is a core concern, with research addressing fairness, transparency, and accountability. Methods are designed to ensure models can be trusted, particularly in high-stakes environments. Applications are prominent in medicine and healthcare, where data-driven insights inform diagnosis and prognosis, and in broader decision-making settings where transparency is essential, such as defence contexts.
Computer vision and multimodal perception
Research in computer vision combines geometric modelling with machine learning to interpret visual information. This includes work on multi-view geometry, 3D reconstruction, object recognition, scene understanding, and visual tracking, as well as their mathematical foundations. Moreover, problems such as image captioning and visual question answering are studied from both multimodal modelling and learning perspectives. Our research also covers novel sensing methods and methods for alternative navigation solutions for autonomous systems. These methods are central to applications in autonomous systems, where perception supports navigation and interaction, and to defence systems and medical imaging, where accurate interpretation of visual data is critical.
Natural language processing and data fusion
Research in natural language processing focuses on extracting meaning from text using learning-based methods. Information extraction challenges in specialised domains such as clinical and biomedical text and in defence, where structured knowledge must be derived from unstructured documents, drive our work. This work is closely connected to information fusion, in which multiple data sources, such as textual, structured, image, video, audio, and various sensor data, are integrated into combined representations to support decision-making. These methods are applied in healthcare to combine patient records and clinical reports to make predictions and synthesise medical reports, and in broader contexts such as defence and intelligence analysis, where diverse data streams must be synthesised into actionable insights.
Knowledge representation and hybrid AI
Research in knowledge representation and hybrid AI focuses on combining symbolic reasoning with data-driven learning to build systems that learn from data and reason with structured knowledge. Ontologies play a central role by providing formal representations of domain knowledge and enabling ontology-based data standards that support consistent interpretation and sharing of information. This work underpins advanced data integration and interoperability among systems, allowing heterogeneous data sources to be combined across complex environments. Such approaches are critical for overcoming data silos in resource industries and engineering, where information is often fragmented across systems and organisations. By linking structured knowledge with machine learning models, hybrid AI enables intelligent decision support systems that operate effectively in domains such as agriculture, manufacturing, and maintenance, supporting tasks like diagnostics, optimisation, and planning while ensuring decisions remain interpretable and grounded in domain expertise.
Robotics, sensing, and embedded intelligence
Research in this area focuses on intelligent systems that interact with the physical world by combining sensing technologies with embedded computation. Work includes remote sensing technologies, signal processing, and methods for interpreting noisy and incomplete data. The research supports the design of systems that operate reliably under real-world constraints. Applications include smart environments, tracking and monitoring systems, and various defence settings.
Reinforcement learning
Research in reinforcement learning examines how agents learn to act through interaction with dynamic, uncertain environments. This includes theoretical work on learning strategies, probabilistic modelling, and sequential decision-making. These methods are directly relevant to autonomous systems, where decisions must be made in real time and adapt to changing conditions. Applications include robotics, control systems, optimisation tasks, and defence scenarios that require adaptive strategies. The research contributes to building systems that can learn from experience and improve over time. Applications also include distributed decision-making in defence contexts, robotics, and autonomous satellite systems.
Multi-agent systems and autonomous decision-making
Research in multi-agent systems examines how multiple agents interact, coordinate, and make decisions in shared and contested environments. This includes formal models of agent behaviour, distributed reasoning, distributed consensus, adaptability, and mechanisms for cooperation and negotiation. These approaches are applied in systems with distributed control and knowledge, such as enterprise systems, autonomous fleets, and organisational workflows. They are particularly relevant to defence scenarios requiring coordination across multiple agents in adversarial settings, as well as to large-scale simulations and distributed software systems.
Human-centred AI and software intelligence
Research in human-centred AI focuses on designing systems that are understandable, usable, and aligned with human needs. This includes work on explanation, decision support, and the modelling of complex physical and cyber systems. In software intelligence, our research develops methods that support system design, diagnostics, and maintenance through formal reasoning and AI techniques. These approaches are used in engineering and manufacturing contexts as decision-support tools, in software engineering to assist developers, to address multifaceted cybersecurity challenges, and in organisational contexts to improve situational awareness, workflows, automation, and decision-making processes, thereby augmenting human ability.
Proven Impact
The School’s AI research delivers impact across a broad range of application domains. In healthcare and medicine, it supports clinical decision-making through interpretable models, the extraction of structured knowledge from clinical records, and the analysis of medical imaging, thereby contributing to improved diagnosis and patient risk assessment. In autonomous systems, research in perception, optimisation, and reinforcement learning underpins capabilities such as visual scene understanding, adaptive control, and real-time decision-making in complex environments. Contributions to defence and security include robust image and video analysis, distributed decision-making using multi-agent systems, multi-source information fusion, and optimisation methods for planning and coordination, all of which support situational awareness and strategic decision-making. In software engineering, advances in explainable AI, formal modelling, and intelligent system analysis enable more reliable system design, development, and maintenance, while data mining and causal learning techniques provide actionable insights into system behaviour and performance across a wide range of situations. These areas and applications demonstrate strength in combining theoretical advances, including causal discovery, geometric vision methods, and evolutionary optimisation, with deployment-driven challenges. This positions its researchers to work effectively in multidisciplinary teams, where addressing complex societal, technical, and theoretical problems requires integrating expertise from AI, domain sciences, and engineering into coherent, high-impact solutions. Together, these themes reflect the School’s focus on rigorous research that connects theory with real-world impact.
