|
Atmospheric Chemistry Modelling DevelopmentPrinciple Investigator: Wuhu Fenghttps://environment.leeds.ac.uk/see/staff/1256/dr-wuhu-feng Project DescriptionOne of the main research topics in atmospheric science is to make fundamental advances in our understanding of climate change, which is one of the great environmental challenges related to changing atmospheric composition, also highlighted as one of the NCAS Science Themes (e.g., Long-term global change). Depletion of the stratospheric ozone layer remains a persistent environmental issue. Ozone depletion in the polar stratosphere is caused by chlorine and bromine species, which are activated by low temperatures. This chlorine and bromine are transported to the stratosphere following the surface emission of ozone-depleting substances (ODSs). This project is to investigate the different roles of different processes (transport and chemistry) on Antarctic ozone hole and Arctic ozone depletion, and how the future ozone changes under differing scenarios.
|
|
Temporal Graph ConvNets for Health RecordsPrinciple Investigator: Samuel Reltonhttps://medicinehealth.leeds.ac.uk/medicine/staff/706/dr-samuel-relton Project DescriptionThis project will develop state-of-the-art techniques for predicting patient outcomes from electronic healthcare records. These will be used to provide clinical decision support to doctors across a range of clinical areas, including cancer, musculoskeletal, and geriatric medicine.
|
|
Multilingual Language TechnologyPrinciple Investigator: Serge Sharoffhttps://ahc.leeds.ac.uk/languages/staff/1137/dr-serge-sharoff Project DescriptionThe field of Digital Humanities is fairly broad. Our research questions are related to using AI tools for getting anything that is expressed through the medium of language, for example:
|
|
Deep Learning and Partial Differential EquationsPrinciple Investigator: He Wanghttp://drhewang.com/ Project DescriptionCombining data-driven methods (especially neural networks) and empirical physics models has sparked interest recently. This kind of method enjoys the explainability of model parameters while being able to quickly predict results, solve inverse problems, etc. However, the two approaches are fundamentally different. Deep learning aims for models with sufficient learning capacity and expressivity, and then is mainly driven by data; empirical physics models employ strong prior knowledge with embedded explainability. The former is usually not explainable, while the latter is not amicable to data. This project explores in several directions how the two distinctive philosophies can be combined and integrated into modelling.
|
|
Centre for Environmental Modelling and ComputationPrinciple Investigator: Mark Richardsonhttps://environment.leeds.ac.uk/see/staff/1500/dr-mark-richardson Project DescriptionCEMAC is a group within the School of Earth and Environment at the University of Leeds. The small team of software development scientists work with research scientists on the computational aspect of their projects. This ranges from tools to manage, process, and visualize data to enhancing the computational techniques within simulation software.Our work on Bede will update and inform the existing software of this research topic.
|
|
SENSEPrinciple Investigator: Anna Hogghttps://environment.leeds.ac.uk/see/staff/1330/dr-anna-e-hogg Project DescriptionThe SENSE Centre for Doctoral training (CDT) is an exciting new centre funded by the Natural Environment Research Council and the UK Space Agency, that will train 50 PhD students to tackle cross-disciplinary environmental problems by applying state-of-the-art data science methods to the deluge of satellite data collected each day. Our graduates are supervised by a consortium of world-leading UK scientists, with topics co-developed with the UK’s most innovative spatial data companies. By training a new generation of industry-experienced satellite data specialists, we are supporting the growing strategic importance of remote sensing within the UK space sector, and enhancing the UK’s profile as an international leader in Earth Observation science.
|
|
Support for UKRI CDT in AI for Medical Diagnosis and CarePrinciple Investigator: David Hogghttps://eps.leeds.ac.uk/computing/staff/84/professor-david-hogg Project DescriptionThe project is supporting PhD students in the UKRI Centre for Doctoral Training in AI for Medical Diagnosis and Care. Many of these students are applying machine learning to explore new ways of diagnosing and treating cancer, often using and combining images, genomic, and molecular data. The projects of our first cohort of nine students span several areas and can be seen in outline on our website (https://ai-medical.leeds.ac.uk/home/student-profiles/). Overall, 50 PhD students will be trained in the CDT over the next five years. The Bede server will provide an important compute-resource to support the work of those CDT students using deep learning methods.
|
|
INSILEX: Computational Precision Medicine for In-Silico Trials of Medical DevicesPrinciple Investigator: Alejandro Frangihttps://eps.leeds.ac.uk/computing/staff/1535/professor-alejandro-f-frangi Project DescriptionINSILEX envisions a paradigm shift in MD innovation where quantitative sciences are exploited to carefully engineer MD designs, explicitly optimise clinical outcomes, and thoroughly test side effects before being marketed. In-silico clinical trials (ISCT) are essentially computer-based MD trials performed on populations of virtual patients. They use computer models/simulations to conceive, develop, and assess devices with the intended clinical outcome explicitly optimised from the outset (a priori) instead of tested on humans (a-posteriori). This will include testing for potential risks to patients (side effects) exhaustively exploring in-silico for MD failure modes and operational uncertainties before being tested in live clinical trials. Advanced computer modelling will prove useful to predict how a device behaves when deployed across the general population or when used in new scenarios, outreaching the primary prescriptions (device repurposing), helping to benefit the widest possible target group without unintended consequences of side effects and device interactions.
|
|
DEEPVOLCPrinciple Investigator: Andy Hooperhttps://environment.leeds.ac.uk/staff/1334/professor-andy-hooper Project DescriptionDEEPVOLC will help to forecast activity at volcanoes by applying advances in artificial intelligence to transformative new geodetic datasets.Some 200 million people live within 30 km of a volcano but accurate forecasting of volcanic activity is problematic. Usually, it relies on human expertise at individual volcanoes, but volcanoes often behave in unexpected ways not previously observed at that location. An additional complication is that most volcanoes are not instrumented. A key indicator of volcanic activity is deformation of a volcano's surface due to magma migrating beneath, and recent advances in satellite monitoring now allow us to monitor this deformation worldwide. DEEPVOLC will apply the latest deep learning algorithms to the satellite data to combine knowledge from all volcanoes that have been active in the satellite-monitoring era. This will enable us to use knowledge of how volcanoes behave globally to identify deformation at volcanoes locally and forecast how it will evolve. Through working with volcano observatories throughout the project, we will deliver tools that can be used to aid in the forecasting of volcanic activity.
|
|
3D image reconstruction using generative AIPrinciple Investigator: Arash Rabbanihttps://eps.leeds.ac.uk/computing/staff/11422/dr-arash-rabbani Project DescriptionWe are planning to study 3D porous material micro-structures from 2D images. This project will benefit environmental and geological studies to understand micro-structures of geological materials from limited available data. We will be using generative artificial intelligence to translate 2D images into 3D.
|
|
Principle Investigator: Zheng Wang https://eps.leeds.ac.uk/computing/staff/6452/professor-zheng-wang Project DescriptionDeep neural networks (DNNs) with billion- and trillion-scale parameters have demonstrated impressive performance in solving many tasks. Unfortunately, training a billion-scale DNN is out of the reach of many data scientists because it requires high-performance GPU servers that are too expensive to purchase and maintain. This project will find ways at the compilers and operating systems level to make the training and use of large deep learning models accessible to mainstream data scientists by reducing their computational resource requirements.
|
|
Clinical Narrative Retrieval based on Deep Learning Approach that utilizes Semantic FeaturesPrinciple Investigator: Lena AlMutairhttps://eps.leeds.ac.uk/computing/pgr/7616/lena-almutair Project DescriptionThis thesis aims to develop an automated clinical information retrieval system that can extract relevant information from unstructured electronic health record (EHR) text data to improve the efficiency and accuracy of clinical decision-making. Specifically, this research will focus on developing a context-enhanced network that can effectively and semantically search EHRs and retrieve relevant clinical notes given a practitioner query.
|
|
Physics informed deep learning for groundwater predictionPrinciple Investigator: Xiaohui Chenhttps://eps.leeds.ac.uk/civil-engineering/staff/756/dr-xiaohui-chen Project DescriptionThe project "Physics Informed Deep Learning for Groundwater Prediction" aims to revolutionize groundwater prediction by combining the power of deep learning with the physical principles governing groundwater flow. Groundwater is a vital natural resource that plays a crucial role in water supply, ecosystem health, and agricultural productivity. However, accurately predicting groundwater levels and flow patterns is a challenging task due to the complex and nonlinear nature of subsurface hydrological processes. This project proposes to develop an innovative approach that integrates physics-based constraints into deep learning algorithms, enabling the extraction of valuable insights from limited and noisy data. By leveraging the strengths of both data-driven deep learning and physics-based modeling, this project seeks to enhance our understanding of groundwater systems, improve prediction accuracy, and support sustainable management of this invaluable resource.
|
|
Physics Informed Deep Learning for Inverse Problems in Solid MechanicsPrinciple Investigator: Peter Jimackhttps://eps.leeds.ac.uk/computing/staff/82/professor-peter-jimack Project DescriptionOur work seeks to allow identification of mechanical properties of a material (Young's modulus and Poisson's ratio) based upon sparse observation of displacements (e.g. using sensors) given known boundary data. By means of physics-informed neural networks we will reconstruct the full displacement field in a manner that ensures physical laws are satisfied and our observations are respected. This deep learning approach simultaneously solves the inverse problem to identify the (not generally homogeneous) mechanical properties.
|
|
Computational endoscopy, surgery, pathology and low-cost devicesPrinciple Investigator: Sharib Alihttps://eps.leeds.ac.uk/computing/staff/11465/dr-sharib-ali Project DescriptionMy group works on medical image analysis. We work on endoscopic image analysis and surgical computer vision techniques. We are looking at multimodal datasets. My group (four PhD students) is working on developing machine learning (in particular deep learning) models for helping clinical colleagues understand the variables in patient treatment and diagnosis.
|
|
Brain-inspired reinforcement LearningPrinciple Investigator: Jian Liuhttps://eps.leeds.ac.uk/computing/staff/9216/dr-jian-liu Project DescriptionIn everyday life, we perceive the surrounding environment, convert sensory information (the colour of traffic lights) into neural activity in our brain, and generate a series of behaviours (to stop or move forward). This whole decision-making system in the brain includes a series of process from sensory perception to motor action. Similar to this, an intelligent machine system is demanding for such a brain-like behaviour to handle both input sensory signals and output decision reactions. It is desirable that we could mimic brain functions and build an intelligent machine system for better implementing practical tasks. This project aims to develop a novel functional framework of brain-inspired reinforcement learning for machine visual and decision making system for the fast and accurate manipulation of autonomous robots.
|
|
Advanced Crystal Shape Descriptors for Precision Particulate Design, Characterisation and ProcessingPrinciple Investigator: Thomas Hazlehursthttps://eps.leeds.ac.uk/staff/835/dr-tom-hazlehurst Project DescriptionNew technology to aid in the crystallisation of precision crystals with pre -defined surface properties will be developed. To do this, we will apply machine learning based upon crystallographic modelling of crystal morphology (forward engineering) to map from 2D in-process temporal microscopy data back to a description of a crystal’s 3D shape (reverse engineering) and, through this, to its functional surface properties. This digital crystal passport will enable the design and control of more efficient and agile manufacturing processes for crystalline fine chemicals, such as pharmaceuticals, agrochemicals, and speciality additives, delivering precision crystals with a much tighter specification in terms of their size and shape than is currently feasible, hence resulting in products having more consistency, less variability, higher quality, and enhanced performance.
|
|
Semi-quantum calculations of magnetic materialsPrinciple Investigator: Joseph Barkerhttps://eps.leeds.ac.uk/physics/staff/5729/dr-joseph-barker Project DescriptionWe are investigating just how much quantum behaviour must be added to simulations of magnetic materials to be able to predict their properties. Too much detail of the quantum nature makes simulations very computationally expensive, but not enough means the results are incorrect.
|
|
Computational studies of membrane proteinsPrinciple Investigator: Antreas Kallihttps://medicinehealth.leeds.ac.uk/medicine/staff/485/dr-antreas-kalli Project DescriptionAs blood is pumped around the body blood, flow puts a lot of pressure on the cells which form blood vessels. The body must therefore sense changes in blood flow so that it can adapt to changes in this pressure and maintain a functioning cardiovascular system. PIEZO1 sits in the membrane of these cells which line the blood vessels and can detect the pressure caused by the blood flow, however how it does this is not well understood. Using computational methods to simulate the pressure applied to these cells, this project aims to understand the changes within PIEZO1 that enable it to sense changes in blood flow and blood pressure.
|
|
Investigating protein-lipid interactions with molecular simulations and machine learningPrinciple Investigator: Antreas Kallihttps://medicinehealth.leeds.ac.uk/medicine/staff/485/dr-antreas-kalli Project DescriptionThe vast majority of drugs target proteins that are embedded in cell membranes. Cell membranes are made of a complex mixture of oily molecules called lipids. This hydrophobic (water-hating) lipid environment makes membrane proteins very challenging to study experimentally. It's also clear that binding sites for particular lipids on membrane proteins are used to regulate the protein's structure and activity, and are also the sites at which many drugs interact.Molecular dynamics simulations allow us to realistically model membrane proteins in their lipid environment, filling the gaps in knowledge that exist due to current experimental limitations. In this project, we are performing these simulations at a large scale, to build a database of lipid binding sites for hundreds of membrane proteins. This allows us not only to identify these important binding sites, but also to understand structural features on the protein that create them. We will then use machine learning on this dataset to create a tool which can quickly predict likely lipid binding sites without having to run computationally demanding simulations.
|
|
A New Order of Liquids: Ferroelectric Nematic Liquid CrystalsPrinciple Investigator: Richard Mandlehttps://eps.leeds.ac.uk/staff/8300/dr-richard-mandle Project DescriptionFerroelectric nematic liquid crystals combine the properties of fluids with the ferroelectric and non-linear optical properties of solids. Currently only a handful of materials are known to exist, and general design principles for new materials are elusive. Large scale atomistic molecular dynamics simulations can be used to guide synthesis by screening candidate materials in silico, allowing experimental investigation to be targeted towards materials of particular interest.
|
|
Parallel Peak PruningPrinciple Investigator: Hamish Carrhttps://eps.leeds.ac.uk/computing/staff/499/professor-hamish-carr Project DescriptionAs data analysis scales to the exabyte, development focuses on analytic code that operates in situ - i.e. on the same machine as the computation. For this purpose, existing serial topological algorithms such as the contour tree break down, and custom algorithms invoking both OpenMP and MPI style parallelism are needed. Development of such algorithms, therefore, relies on access to machines such as Bede which exhibit this architecture.
|
|
Adopting green solvents through prediction reaction outcomes with AI/Machine LearningPrinciple Investigator: Bao Nguyenhttps://eps.leeds.ac.uk/chemistry/staff/4205/prof-bao-nguyen Project DescriptionThe switch from traditional organic solvents, many of which are hazardous, volatile or non-sustainable, to modern green solvents is one of the key sustainability objectives in High Value Chemical Manufacture. Currently, the use of green solvents is often explored at the process development stage, instead of the discovery stage. This necessitates re-optimisation of processes, due to changes in yield, selectivity, impurity profile, and purification. These lead to longer development time, cost, and additional uncertainty. On the other hand, selecting the right solvent early may enhance chemoselectivity, avoid additional reaction steps, and simplify purification of the products. Predicting these changes is an important underpinning capability for wider adaptation of green solvents in manufacturing. Unfortunately, the scarcity of reaction data in green solvents is a key obstacle in developing this capability. Thus, there is an urgent need for ML models that predict reactivity in green solvents based on available data in traditional solvents. In addition to addressing the short time-scale of early-stage process development, these will increase the confidence in utilising green solvents at the discovery stage, supports sophisticated synthetic routes planning tools which takes into account side products, impurity and purification methods, and act as valuable regulatory tools for assessing hazardous impurities.
|
|
Physics-based machine learning in the CDT in Fluid DynamicsPrinciple Investigator: Peter Kane Jimackhttps://eps.leeds.ac.uk/computing/staff/82/professor-peter-jimack Project DescriptionHistorically, we have attempted to solve complex problems in fluid dynamics - such as forecasting the weather, designing vehicles that have low aerodynamic drag or understanding the blood flow through a patient's beating heart - through the use of computational algorithms that aim to accurately approximate the local physics at each point in space and time. Recent developments in machine learning allow the possibility of using data to generate models that run significantly faster - but often with less reliability or understandability. In this work we seek to develop new methods, so-called physics-informed machine learning, which contain features from both approaches to ensure that data is enhanced by knowledge of the underlying physics.
|