Navigation
Applications of Artificial Intelligence
In January 2023, I joined the University of Hull as a lecturer (assistant professor) at the Centre of Excellence for Data Science, Artificial Intelligence, and Modelling. Influenced by this intellectual environment, I began to work in applied artificial intelligence in parallel to systems research.
Real-Time 3D Reconstruction from 2D Human Images
I am the secondary supervisor of Gulraiz Khan, a PhD candidate. He developed a deep learning framework to predict the facial landmarks, ages, and genders of human face images. It has a backbone model (ResNet-18) for feature extraction. It is linked to three branches of lightweight neural networks, which share the convolutional feature maps produced by the backbone model in response to an image. They predict the image's facial landmarks, age, and gender respectively. He is now solving the problem of real-time, identity-aware, and clothing-aware 3D reconstruction.
Platelet Proteomes
Platelets are small blood cells that play a key role in blood clotting (haemostasis). On the other hand, uncontrolled platelet activation could result in thrombosis, the formation of a blood clot inside a blood vessel, which obstructs the flow of blood through the circulatory system. Multiple stimulatory and inhibitory signalling pathways interact within a platelet to enable it to adapt to a changing extracellular environment. For example, its proteome may change in response to oxidative stresses. Dr Giordano Pula (Hull York Medical School) and I are interested in identifying such changes and then interpreting them in functional terms.
In diabetics, platelets are known to be hyper-responsive and responsible for cardiovascular diseases such as myocardial infarction, stroke, or peripheral tissue ischemia. In the autumn trimester in 2023, Shaon Biswas conducted his MSc dissertation research (AI and Data Science) under our supervision at the University of Hull. He used the Enrichr library, the KEGG database, and the UniProt database to identify the platelet proteins associated with cardiovascular complications in a proteomic dataset pertaining to 23 diabetic (type 2) patients and 24 controls. Using this list of proteins as his reference, he combined standard differential expression analysis with protein-protein network analysis (STRING database), feature importance analysis (random forest classification), clustering, principal component analysis, and association rule learning to identify novel platelet protein biomarkers of cardiovascular complications in diabetics.
MSc Projects
Our MSc degree programme in artificial intelligence and data science includes a dissertation project. Throughout the academic year, I propose and/or supervise these projects. Sometimes, our students design their own projects. Sometimes, we collaborate with external partners.
My students and I have developed applications of artificial intelligence in areas such as healthcare, sport science, platelet proteomics, cancer transcriptomics, fintech, medical image classification, patient record classification, retrieval-augmented generation, optical character recognition, image captioning, multimodal sentiment analysis, dysarthric speech recognition, and optimisation.
In the summer trimester of 2023, I supervised 14 projects, including collaborative projects with the Academic Vascular Surgical Unit (AVSU) of Hull Royal Infirmary, a tertiary teaching hospital run by an NHS hospital trust. Enoch Appiah built an image processing pipeline and a CNN-LSTM model to digitise handwritten clinical notes. He won the Best MSc Research Project Prize. Folakemi Fatiregun built an AI system to allocate hospital beds. Partly on the strength of her impressive dissertation, she secured a full-time position as an information analyst at the same NHS trust.
In the autumn trimester of 2023, I supervised 12 projects. Mustabshira Zia built an automatic speech recognition system for dysarthric people. Esther Chukwuezi experimented with diverse dimensionality reduction techniques to optimise risk prediction based on cardiopulmonary exercise testing data about people with heart failure.
In the summer trimester of 2024, I supervised three projects, including two collaborative projects with Spencer Group, a civil engineering company in Hull. Within the framework of retrieval-augmented generation, Duy Thanh Do used the company's database of emails and documents to optimise the outputs of large language models, giving them the ability to produce summaries of project milestones. Working with the same database, Olalekan Akindele used knowledge graphs to enhance retrieval-augmented generation, turning large language models into chatbots. Olalekan won the Best MSc Research Project Prize.
