|
Networks of Cardiovascular Digital TwinsPrinciple Investigator: Richard Claytonhttps://www.sheffield.ac.uk/dcs/people/academic/richard-clayton Project DescriptionDigital twins are a representation of an actual object, such as a wind turbine, stored in a computer and used to inform how the object is used and maintained. This project aims to build and deploy digital twins of the heart and circulatory system of individual patients with a heart disease called pulmonary hypertension. We will develop methods to create a digital twin of newly diagnosed patients, and to update digital twins with new information each time the patient has a scan or a visit to their GP. The digital twin will be used to forecast how the disease will develop in an individual patient, and this information will be used to guide treatment and optimise care such as anticipating the need for hospital visits.
|
|
Deep learning segmentation of MR and microscopy imagesPrinciple Investigator: Alberto Biancardihttps://www.sheffield.ac.uk/smph/people/clinical-medicine/alberto-biancardi Project DescriptionAccurate segmentation of MR and microscopy images is a challenging and demanding task, if performed manually; moreover, fatigue can lower markers' performance, requiring additional breaks and reducing throughput. We will pursue an automated segmentation of MR images to achieve very high accuracy while processing large amounts of data for our current studies.
|
|
Multimodal models of Natural Language Processing for Non-literal languagePrinciple Investigator: Marvin Bohua Penghttps://www.sheffield.ac.uk/dcs/people/research-staff/marvin-bohua-peng Project DescriptionOur project aims to address the limitations of modern language models in processing idiomatic expressions, which are crucial for achieving fluency and accurate natural language understanding in all languages. Idiomaticity poses unique challenges for NLU tasks, including the need for background knowledge, chain of thought reasoning for metaphor, and the inclusion of visual information.To tackle these challenges, our project will focus on developing a more explainable approach that can incorporate the contextual visual knowledge of an idiom. We will propose a general methodology for infusing textual and visual knowledge together.
|
|
Agent-based modelling of actin polymerisationPrinciple Investigator: Mike Williamsonhttps://www.sheffield.ac.uk/biosciences/academic-staff/mike-williamson Project DescriptionThe overall aim of the project is to simulate the processes involved in the nucleation and polymerisation of actin when it forms endocytic patches in yeast. This process involves the binding (and unbinding) of a number of proteins, which compete with each other for binding sites. The coding of the project uses an environment callled FLAME GPU in a novel application, which has already demonstrated that using a computational approach in conjunction with experimental measurements brings new insight that could not be delivered by either approach in isolation. The computer application requires a GPU architecture as an essential component.
|
|
Natural Language ProcessingPrinciple Investigator: Chenghua Linhttps://www.sheffield.ac.uk/dcs/people/academic/chenghua-lin Project DescriptionMachine learning, natural language processing, data, and text mining. The development (& testing) of algorithms and models for sentiment analysis, text summarisation, natural language generation, and cognitive-inspired context learning.
|
|
vera.ai: VERification Assisted by Artificial IntelligencePrinciple Investigator: Xingyi Songhttps://www.sheffield.ac.uk/dcs/people/academic/xingyi-song Project Descriptionvera.ai is a Horizon Europe project aimed at basic and applied research of AI methods to fight false information. The vera.ai project focuses on textual, multilingual, and multimodal content, and puts strong emphasis on context and inter-content relationships.
|
|
Building large knowledge graphs using linked dataPrinciple Investigator: Ziqi Zhanghttps://www.sheffield.ac.uk/is/people/academic/ziqi-zhang Project DescriptionThe project aims to develop novel language resources through exploiting the gigantic HTML-embedded structured data on the Web, currently made available through the Web Data Commons (http://webdatacommons.org/) project. Such language resources will be mined to develop a ‘proof-of-concept’ Knowledge Graph, which is a knowledge representation used by modern search engines and also in many applications for knowledge discovery. Despite the rapid growth of such structured data on the Web and the continued effort in harvesting and releasing such data to the community under the WDC project for the last 10 years, there is still a lack of research into using such data for different computational tasks. We believe this will be the first but an important step towards unlocking the potential of this data, thus fostering community effort in this direction of research. The resources created from this project will also benefit the wider NLP, Semantic Web, Machine Learning, and Data Mining communities in general.
|
|
The competing dynamics of damage and sequence in supercoiled DNAPrinciple Investigator: Tim Craggshttps://craggs-lab.com/ Project DescriptionDNA damage is repaired with varying efficiency across the genome. Is this linked to the accessibility of the damage itself? When supercoiled, DNA often forms a structure called a plectoneme that protrudes out of the densely packed genome. This leads to the hypothesis that damage may locate to the apices of the plectoneme, otherwise known as pinning the tip. If damage does indeed do this, it would reduce the search space for damage-recognizing proteins within DNA repair pathways, as well as provide bent or flipped structures for their binding. Sequences of high AT content also pin plectonemes and so by simulating two sequences containing different amounts of AT, and therefore, varying plectoneme pinning propensities, we can study why some regions of the genome are repaired less efficiently than others.
|
|
Machine learning models for turbulent wakes over fractalsPrinciple Investigator: Yi Lihttp://yi-li.staff.shef.ac.uk/index.html Project Description"Simulating turbulent fluid flows is difficult and yet many real-life situations require the ability to do so. For example, there is a need to understand how the air flow past a building affects how the building oscillates, or understand how to shape a car to become more carbon-friendly, or understand how air pollutants in cities are dispersed by the wind. Yet, currently, the best method for predicting the flows in these cases requires the use of supercomputers and large amounts of storage space.To reduce the computational demand, this project aims to use machine learning models to make predictions of the key features of such flows. The hope is that these models can help researchers to understand how the turbulent fluid acts in these scenarios and advise policy makers or guide aerodynamic engineering designs."
|
|
Understanding Human from Images with Deep LearningPrinciple Investigator: Zhixiang Chenhttps://www.sheffield.ac.uk/dcs/people/academic/zhixiang-chen Project DescriptionThis project aims to develop deep learning models to understand human from images. This includes human (body and hand) pose estimation, detecting human and object interaction, and human body/face/hand reconstruction. This project will study the state-of-the-art deep learning models including both the Convolutional Neural Networks and the Transformer based models. The output of this project will benefit the development of downstream applications like human machine interaction, telepresence and VR/AR.
|