School of Science & Technology Doctoral Studentships

The introduction of digital technologies into personal money management has unintended and sometimes adverse consequences for citizens. These exacerbate financial exclusion and affect disproportionally those who struggle financially or find themselves in vulnerable circumstances. Despite these problems, the discourse surrounding financial technologies continues to be narrowly focused on their purported benefits, with little attention being paid to their shortcomings and, most importantly, to how they could be addressed.

This PhD will research how the downsides of digitising personal finance could be tackled through design. It will do so through participatory design methods and in collaboration with autistic adults.

Participatory design is an approach to technology design that empowers citizens to actively shape and influence the technology-making process. Autistic adults are a group that has been systematically excluded by technology makers and financial service providers. Given the challenging life and financial circumstances of autistic adults, it is important to assess whether existing financial technologies and services are suitable for their needs and preferences, and whether they are resulting in stigmatising or discriminatory treatment.

In addition, collaborating with autistic adults will yield new, diverse, and innovative perspectives that will help researchers critically assess the current state of financial technologies and push the boundaries of their design.

This project will develop a common digital infrastructure to have a Digital Twin (DT) of both floating and monopile offshore wind turbines. A DT is a cyber-physical system that must represent physical reality at a level of accuracy suited to its purposes. The extent of realism depends on three essentials: modelling, data and visualisation.

In this context, the project will first develop a simulation engine for a realistic numerical representation of the physical behaviour of the assets, combining Finite Elements Analysis (FEA) and multi-physics environments (e.g., wind, waves, earthquakes, etc).

Second, a modern flexible, modular and scalable software architecture will be developed to establish the DT virtual environment, to host and interact with simulation engines.

Third, the DT will be used as a synthetic simulator to produce real-life sensor-like data, of controlled damage and undamaged stages of the assets, for data and Artificial Intelligence (AI) driven solutions for structural health monitoring and damage detection.

AI systems, and in particular Deep Neural Networks have obtained impressive results, from defeating professional players at games to the recognition of protein 3D structures. However, they are restricted to specific tasks and domains. In technical terms, they are good at interpolating data, but not at classifying and predicting data dissimilar to their training datasets. That is, AI does not generalize and transfer knowledge.

Such deficit poses, in turn, the question of explainability. Since AI systems don’t show coherent behaviour across comparable domains and tasks, it is difficult to interpret how they operate and thus account for their results.

The project aims at answering the following questions:

1. Can AI systems that use categorical structures improve generalization and facilitate knowledge transfer across heterogeneous datasets?
2. Can AI systems with richer representations beyond sets and equivalence relations recognize similarities across different domains and tasks?
3. Can such systems be a path towards Artificial General Intelligence?
4. Can we build explainable AI systems based on the formal foundations of category theory?
5. Can category theory serve as a tool to analyse and interpret the operations of AI systems?
6. Can category theory be instrumental in building accountable and responsible AI systems?

Sophisticated chatbots are truly disruptive technologies within an educational context – e.g. OpenAI's ChatGPT has caused a great deal of concern within Higher Education, particularly in terms of their potential use to aid students during the assessments.

However, they also offer opportunities in terms of enhancing learning assistance, particularly for unsupervised learning within an asynchronous online course. A student can interrogate a chatbot to help correction of erroneous software code (across many programming languages), including explanation; to obtain sophisticated answers to fairly standard textbook questions; to obtain rudimentary answers to complex problem-based questions. Current chatbot technologies are far more powerful than previously developed technologies, noting a long history of AI-based reasoning engines, going back to domain-specific expert systems in the '80s.

As well as answering questions to providing learning assistance, sophisticated chatbots can also provide questions to students. Within the context of a particular module, targeted personalised questions can be asked to understand a student's learning progression – both in terms of knowledge and skills. For example, if learning programming a student can be challenged to use various language elements correctly, or to construct solutions to problems.

This research project will seek to understand and enhance the capability of chatbot technologies within both the learning and formative assessment within online modules. It will explore how a personalised learning experience can be built within the context of online education. Inter-disciplinary aspects will be considered, including AI, ontology, pedagogy and ethics.

Nowadays, thousands of IoT devices are interconnected and subject to several attacks. For example, these devices can be compromised and act as botnets in large-scale D/DoS attacks or as eavesdroppers that affect citizens' privacy.

This project aims to investigate, design, and develop an authentication mechanism that continuously and intelligently authenticates IoT/smart devices, which can be compromised in time and act as botnets or eavesdroppers.

The authentication mechanism will consider the device (hardware/software), network traffic and behavioural characteristics of IoT devices with the help of learning algorithms and other mathematical approaches. The learning algorithms will assist in building a device profile and observe any anomaly behaviour of IoT devices, signifying potential malicious activity. New mathematical directions (e.g., use Bent functions/Diophantine equations) will be explored for their suitability for creating unique device fingerprints.

Misinformation, especially on social media, is one of the greatest online harms of our time and is blamed for many social ills (e.g., vaccine hesitancy, election manipulation, and political unrest). People often do not actively seek misinformation; they passively and serendipitously encounter it (e.g., on their social media feeds). However, the role of serendipity in spreading misinformation on social media has not previously been examined.

This PhD will investigate how and why social media users serendipitously encounter, engage and disengage with this potentially harmful content and its influence on their views and behaviour. It will feed this understanding of the ‘sour’ side of serendipitous information encounters into an explanatory model of the role of serendipity in propagating misinformation on social media and guidelines based on the model to inform government policy, information literacy and/or social media platform design.

This PhD will involve detailed qualitative research (e.g., through a mix of interviews, observations of social media use, and diary studies). There is also scope to (optionally) complement this with quantitative research (e.g., surveys on the prevalence of the problem or experiment-based user studies). This is an opportunity to conduct research of high societal importance and to create a strong research impact beyond academia.

Recent deep learning techniques provide unprecedented accuracy in most medical image analysis tasks, but they typically require large image datasets annotated by medical experts. Studies suggest however that expert radiologists can provide suboptimal annotations in up to 30% of the scans (e.g., labelling a healthy image as pathological or incorrectly localising a lesion).

In the presence of these "noisy labels", state-of-the-art techniques:

1. fail to produce accurate models (due to noisy training samples);
2. cannot be reliably evaluated (due to noisy test samples).

This project will focus on designing and implementing novel noise-robust learning (NRL) techniques for medical image analysis. These techniques will likely build upon recent self-supervision and multi-task learning methods. The candidate will focus both on classification and segmentation tasks, using both public and clinical medical image datasets provided by project partners (e.g., the University of Chicago). The candidate will also explore how NRL techniques can be used in active learning scenarios to identify a selection of wrongly labelled samples to be relabelled by a clinical expert.

The developed approaches will have a sizeable contribution on the deployment of deep learning models in the clinic in human-in-the-loop settings.

The centrepiece of the project will be the investigation of early detection of SARS-Cov-2 viruses in clinical conditions, from chest CT scans and x-rays. This project will develop new solutions for segmenting lesions and predicting COVID-19 and other types of pneumonia from chest CT scans using an advanced deep-learning methodology.

We are now moving to an envisioned future power system which is dominated by environment-friendly distributed resources that are highly intermittent and variable. The proliferation of renewable-based generation requires the use of storage devices to maintain adequate reliability levels. However, energy storage is only useful if combined with intelligent algorithms that determine when and how much to charge or discharge the finite-capacity battery.

As part of this studentship, the use of storage along with renewable generation to operate a highly renewable power system will be investigated. More specifically, the promising Carnot Battery technology will be studied where excess renewable energy is used to increase the temperature of low-grade heat which is stored for later release and conversion to electrical power when required.

This project aims to make a step-change to the current design practices of offshore wind turbines (OWTs) used for renewable wind energy generation, by reducing the constructional steel usage in next-gen multi-megawatt turbines (≥15MW) while ensuring their resilience to environmental loads. This is pursued by developing a novel compound optimal design protocol for OWTs fitted with dynamic vibration absorbers (DVAs), in which the OWT tower and the supporting structure are sized for minimum weight together with the optimal tuning of DVAs to meet all design constraints.

The project will tackle the challenges of meeting fatigue and ultimate limit states in large-scale OWTs with ≥200m height under site-specific complex loads by developing novel computationally efficient machine learning and physics-based surrogate models, and by migrating the most recent DVAs from wind-sensitive tall buildings applications.

Thermoacoustic instability is a highly unwanted phenomenon produced by the two-way coupling between flames and acoustic waves. It leads to pernicious oscillations in the combustors of gas turbines, aero engines, and rocket engines which may result in structural damage, reduced energy conversion efficiency, and increased emissions of harmful by-products. The combustion of carbon-free fuels, such as hydrogen, is significantly different to traditional fuels, making them more susceptible to instability. Carbon-free fuels are expected to play a critical role in the decarbonisation of combustion systems and, therefore, suppressing thermoacoustic instability is crucial for a successful transition towards a net-zero economy.

A common technique to reduce the propensity to thermoacoustic instability is the use of perforated plates due to their ability to damp acoustic fluctuations. In this project, we will investigate the acoustic response of perforated plates using a combination of high-fidelity numerical simulations, cost-effective numerical tools, and experiments. This study will inform the design of many engineering systems involving perforations, including the combustors of gas turbines and rockets, leading to the design of more efficient and safer combustion systems.

Let's start with a big question: can we get a person's whole health profile continuously with ease? For example, measure glucose without a prick, obtain blood pressure without a cuff, detect heart arrhythmias, screen for cardiovascular disease biomarkers, and understand skin hydration levels without chunky devices. Photoplethysmogram (PPG) and its ability to quantify and monitor heart rate variability (HRV) have made smartwatches pervasive. However, PPG's full potential for quantifying blood pressure (BP), and other biomarkers relating to our healthcare and well-being and its full capacity for metabolic sequencing are still under exploration.

For example, various studies have used PPG to measure BP, but the results have been inconsistent. The discrepancy in results may be due to the use of different PPG sensors, signal processing methods, and algorithms. There is a great need to develop new methods using PPG that can accurately measure blood pressure and overcome the limitations of the traditional method. We, therefore, propose to conduct this research to establish a ground truth established by clinical experts. Furthermore, to develop a set of advanced signal processing and machine learning analytical pipelines to unlock the potential of using PPG for continuous blood pressure measurement using wearable devices.

A significant portion of neonates admitted into intensive care receive intravenous (IV) fluid therapy, which often results in fluid overload signalled by the consequent formation of generalised oedema. Complications of IV fluid therapy can lead to mortality and significant morbidity for the patient. At present, assessing if the correct amount of fluid has been given is determined using basic visual assessment, together with documenting weight and fluid intake and output. This technique is limited in various ways and does not prevent cases of generalised oedema. It is also infrequently used as the baby is often too clinically unstable to move around and be weighed daily.

This project proposes the first proof-of-principle of a novel, contact-free method for precise and continuous monitoring of fluid balance using hyperspectral imaging in the Near Infrared region (NIR HCI). The research is in collaboration with a leading Children's hospital (Great Ormond Street Hospital for Children (GOSH) and Ulster University. The project aims to utilise this technique to develop hyperspectral fluid distribution maps in critically ill neonates, as well as build quantitative models for the accurate determination of fluid balance.

To bring Internet-grade security to billions of IoT devices, IoT devices must move away from pre-shared keys to digital certificates. New and proposed standards enable IoT devices to implement more lightweight solutions for application layer security, offering real end-to-end security also in the presence of proxies. Using Object Security for Constrained RESTful Environments (OSCORE), IoT devices can communicate securely in a standardized manner using application layer security, which allows messages to traverse proxies to offer end-to-end security.

OSCORE together with Ephemeral Diffie-Hellman over COSE (EDHOC) for key establishment, have the potential to offer lightweight solutions for establishing secure sessions. Hence there is a need for a lightweight and secure certificate enrolment protocol using EDHOC and OSCORE protocols that can traverse proxies without breaking end-to-end security. Combining EDHOC with certificate enrolment is an active area of research and it provides not only authenticated encryption but also provides less message overhead as compared to certificate enrolment over OSCORE.

The project aims to quantify and model human dynamics in decentralised socio-technical systems.

Depending on the common interest of the candidate and the supervisor, it will focus either on the dynamics of information production and consumption on social networks (misinformation, polarisation, etc) or on the analysis of cryptocurrency/web3 ecosystems (crypto transaction networks, NFT market, etc).

The project will involve the analysis of large datasets, statistical analysis and, whenever possible, mathematical modelling. In all cases, it will be developed in collaboration with top research institutions, and major partners in the private and/or public sectors.

Evolutionary game theory describes a process of Darwinian selection of traits or behaviours of individuals. A common assumption in those models is the timescale separation between selection shaping the strategy frequencies and ecology shaping the population size. This is not always realistic, and building on results by Argasinski, Broom and others, we shall consider a framework where games are embedded in the population dynamics.

Eco-evolutionary processes incorporating births and deaths resulting from interactions between different strategists will be studied. Such strategies can include heritable traits or social behaviour with payoffs interpreted in terms of fertility and mortality. We will generalise the work of Argasinski and Broom, the eco-evolutionary stability conditions, to more than two strategies. This will lead to new mathematical and biological insights.

Qualitative analysis of our results in terms of the geometry of phase portraits can reveal important information such as the evolutionary stability of critical points.

The Hochschild cohomology of a finite group algebra is a subtle invariant describing structural properties of blocks of finite groups, closely related to some of the iconic conjectures that drive modular representation theory. This is a graded algebra with some higher structures, one of which is the Lie algebra structure of HH1(A). Empirically, this Lie algebra structure seems to preserve a significant amount of structural information of the original algebra.

The project will aim at solving a number of open conjectures in this area, such as the non-vanishing of HH1(B) of blocks with a nontrivial defect group and the solvability of the Lie algebra HH1(kS_n) of symmetric groups S_n. Furthermore, structural conjectures relate the local structure of blocks to the Lie algebra structure of their Hochschild cohomology.