|
Creativity EnginePrinciple Investigator: Kathryn Garsidehttps://www.ncl.ac.uk/press/articles/archive/2023/05/creativityengine/ Project DescriptionAI text generation in collaboration with Seven Stories - The National Centre for Children's Books.
|
|
NUFEBPrinciple Investigator: Stephen McGoughhttps://www.ncl.ac.uk/computing/staff/profile/stephenmcgough.html Project DescriptionThe modelling of microbial systems is a computationally demanding task – the simulation of just one mm3 of wastewater contains 109 bacteria. In order to model and understand these systems requires immense computational power. We can simulate 1mm3 of wastewater in approximately 24 hours of supercomputer time. However, to understand the emergent properties of such systems, we need to simulate at scales up to 1 litre – far beyond what can be provided by the world’s largest supercomputers. In order to overcome this, we are using Deep Learning Emulators to replace the computationally intense simulations with emulators that can mimic the actions of the simulation. By merging multiple emulators, we will be able to move up to the litre scale. This will require many runs of our simulation for Deep Learning training.
|
|
Computational Medicinal ChemistryPrinciple Investigator: Daniel Colehttps://www.ncl.ac.uk/nes/people/profile/danielcole.html Project DescriptionOur research, funded by a UKRI Future Leaders Fellowship, seeks to model computationally the structure, dynamics, and interactions of molecules at the atomistic level. In particular, our research goal is to derive improved physical models of biomolecular dynamics and their interactions with potential drug molecules and, thereby, to design software to improve the efficiency of pharmaceutical research and development. Access to Bede will substantially increase the speed at which we can build models and test predictions.
|
|
Robust and Interpretable Deep Learning for Biomedical DataPrinciple Investigator: Jaume Bacardithttps://www.ncl.ac.uk/computing/staff/profile/jaumebacardit.html Project DescriptionThis project focuses on the development, training, and validation of deep learning pipelines for the analysis of biomedical data. Deep Learning can obtain state-of-the-art results in many data analysis tasks across a broad range of application areas and data modalities (e.g. tabular data, images, audio, accelerometer data). In their application to biological and biomedical data, special consideration needs to be placed on ensuring the robustness of these models (e.g., in relation to data confounders) and in interpretability (obtaining a precise understanding of how these models make decisions and what knowledge they have captured). This project will focus on the development of computational strategies to tackle these two challenges in biological/biomedical data, e.g., for medical diagnosis/prognosis, forecasting of animal growth data in farms.
|
|
DNS of turbulent lean premixed hydrogen flamesPrinciple Investigator: Andy Aspdenhttps://www.ncl.ac.uk/engineering/staff/profile/andrewaspden.html Project DescriptionHydrogen has the potential to be a carbon-free fuel for transportation, power generation, and long-term energy storage. Hydrogen is a simple fuel, but behaves very differently to conventional fossil fuels. It is crucial to understand particular instabilities characteristic of hydrogen flames to be able to design efficient low-emission combustors. This project focuses on fundamental flame physics to enable to development of turbulent-flame models that can be used in low-cost engineering applications to design the next generation of green combustors.
|
|
Use of deep learning for prediciting therapeutic antibody fragmentsPrinciple Investigator: Anil Wipathttps://www.ncl.ac.uk/computing/staff/profile/anilwipat.html Project DescriptionThe human body fights infections such as COVID-19 by the production of antibodies that bind and kill infectious agents. However, the body takes time to generate these antibodies, and sometimes it isn't fast enough to fight off a disease before damage or death occurs. Antibodies against viruses and bacteria can be prepared in the lab and given to a patient as a drug to help their own immune system. However, determining the composition of this antibody is time-consuming and traditional methods for this task can take months or years. In this project,t we are using deep learning to generate the sequences of new synthetic antibodies that might act as new drugs for treating diseases such as COVID-19. The computer predictions will be tested in the lab, and then further collaborations with medical teams will be sought to follow up on any interesting candidates.
|
|
Zero-Magnets Electric Drive for Electric Vehicles (Z-M Drive)Principle Investigator: Zepeng Liuhttps://www.ncl.ac.uk/engineering/staff/profile/zepengliu.html Project Descriptionwe are developing a large language model (LLM)-based tool to predict the remaining useful life (RUL) of Zero-Magnets Electric Drives. Accurate RUL prediction is crucial for ensuring reliability, safety, and timely maintenance. Training such a model requires significant computational resources, and access to high-performance computing facilities will greatly accelerate this process. The project combines cutting-edge machine learning techniques with real-world engineering challenges to contribute to the future of greener, smarter electric vehicles.
|
|
Correlating neuronal activity and large volume nanoscale imaging using AIPrinciple Investigator: Carmelo Calafiorehttps://rse.ncldata.dev/team/carmelo-calafiore Project Description"Understanding how the brain works - a major driving force for the development of AI - requires knowledge of the wiring of its individual neural circuits. This can be achieved using electron microscopy to image large volumes of brain tissue and map the connections between neurons (synapses) at the nanometre scale. However, while the wiring diagram (connectome) resulting from this effort is necessary to understand the brain, it is not, by itself, sufficient. We also need information about the function of each connection, i.e., how effective each synapse is at information signalling. We aim to complement brain connectomes with a readout of synaptic activity, using new AI-based tools that will allow us to better understand the relationship between brain structure and function in health and disease."
|
|
Neural Network Surrogate Models for Hot Stamping FeasibilityPrinciple Investigator: Jichun Lihttps://www.ncl.ac.uk/computing/staff/profile/jichunli.html Project DescriptionThe automotive and aerospace industries are increasingly using lightweight aluminum alloys to enhance fuel efficiency and reduce emissions. However, forming these materials is challenging due to their limited formability under conventional methods. This project focuses on applying deep learning techniques to predict the manufacturing feasibility of components formed through the advanced HFQ process. By utilizing neural networks trained on large datasets of simulation results, this research aims to develop a tool that can rapidly predict forming outcomes like deformation, thinning, and wrinkling in real-time. This will support engineers in optimizing component designs early in the process, reducing costs and accelerating the adoption of lightweight materials. The N8 HPC resources will be critical for training these complex models on extensive datasets, enhancing prediction accuracy and scalability.
|
|
Unified large vision language modelPrinciple Investigator: Zhuang Shaohttps://scholar.google.com/citations?user=kR-lUmQAAAAJ&hl=zh-CN Project DescriptionIn real-time robotic applications, compound tasks are needed (etc. grab the correct blocks or interact with human beings). However, the modality of information tasks can vary from language to task. Recently, large vision language models have proved effective for many vision tasks, but without too many unified tasks on robotics. This project is to explore a unified large vision language model for many vision tasks to facilitate robotics.
|
|
Organ Quality AssessmentPrinciple Investigator: Robin Nandihttps://rse.ncldata.dev/team/robin-nandi Project DescriptionOrgan Quality Assessment is a multidisciplinary project involving academics, clinicians and software engineers. The aim of this project is to leverage moderm machine learning techniques to generate robust automated predictions of transplant viability from images of candidate organs.
|
|
Multi-modal generation for healthcare applicationsPrinciple Investigator: Shuanglin Lihttps://www.ncl.ac.uk/engineering/research/electrical-electronic-engineering/intelligent-sensing-communication/ Project DescriptionGenerative AI for facial expressive behaviours.
|
|
Numerical Simulation of CPT using PFEM with Application to Offshore GeotechnicsPrinciple Investigator: Gosai Alyamanihttps://research.ncl.ac.uk/ne2g/people/staffprofilegosaialyamani.html Project DescriptionCone Penetration Testing (CPTu):
|
|
Developing Patient-Specific Epilepsy Prediction Model with Longer Forecast Windows Based on Probabilistic ClassificationPrinciple Investigator: Jichun Lihttps://sites.google.com/view/jamesdawson/home Project DescriptionAs one of the most common neurological disorders, epilepsy affects millions of families around the world. In many cases, patients with epilepsy die not because of epilepsy itself, but because of unexpected accidents caused by it, such as falls, burns, and drowning. Thus, we must predict the potential epileptic discharge and take precautions to prevent serious accidents. In this study, we aim to propose a patient-specific epilepsy prediction model based on several state-of-the-art techniques. Graph neural networks will be used for feature extraction, which can better capture the correlation between electrodes. In epilepsy prediction, one challenge is that the inter-subject variability is high, which makes the generalization ability of traditional epilepsy prediction models is poor. To solve the problem, we will transfer the knowledge AI models learned from the whole dataset to the epilepsy prediction task of a specific patient, and then improve the performance of the proposed model. Based on a variety of cutting-edge technologies, we expect to provide new options for epilepsy prediction.
|
|
Machine learning for quantum gasesPrinciple Investigator: Thomas Billamhttps://www.ncl.ac.uk/maths-physics/people/profile/thomasbillam.html Project DescriptionOur project aims to revolutionize the way we understand and study quantum vortices in Bose-Einstein condensates, a unique state of matter that exists at ultra-low temperatures. Using machine learning techniques, we plan to reconstruct the complex topological features of these vortices, which are like tiny tornadoes at the quantum level, from the limited data measurable in typical experiments. Traditional methods have limitations in capturing these intricate structures; machine learning offers a more efficient and accurate approach. Access to high-performance computational facilities will accelerate our research, allowing us to run large-scale simulations and complex machine learning models. The project promises to provide new insights into the fascinating world of quantum gases and their vortex behaviors.
|
|
Machine Learning Interatomic Potentials (MLIPs) for Energy MaterialsPrinciple Investigator: Ioan Bogdan Magdauhttps://www.ncl.ac.uk/nes/people/profile/ioanmagdau.html Project DescriptionThis project will develop machine learning inter-atomic potentials (MLIPs) for molecular materials and organic-inorganic interfaces relevant to energy storage technology and medicinal applications. The field of MLIPs is rapidly evolving, and it is poised to become central to the future of atomistic simulations. The efficiency and accuracy of MLIPs have been thoroughly studied on fixed train/test datasets but the performance of these models in actual molecular dynamics (MD) remains poorly understood. Condensed phase molecular systems are particularly challenging to model owing to a large difference in scale between intra- and inter-molecular interactions. This project will test the stability and accuracy of General-purpose potentials (GPP) on molecular condensed phase applications, explore ways to enhance their accuracy in describing inter-molecular interactions and develop active learning protocols for fine-tuning GPPs for molecular materials.
|
|
Predicting ocean subsurface dynamicsPrinciple Investigator: Matthew Crowehttps://www.ncl.ac.uk/maths-physics/people/profile/matthewcrowe.html Project DescriptionThe upper layer of the ocean is called the 'surface mixed layer' and is responsible for exchanging gases between the ocean and atmosphere and for dissipating much of the energy that enters the ocean due to wind and tides. At present, global ocean models are unable to properly capture these processes due to resolution constraints. This project aims to use machine learning techniques to understand and predict the dynamics of the upper ocean by building models that incorporate these processes. These models would greatly improve the predictive power of the next generation of climate simulations.
|
|
AI for Climate Change ModellingPrinciple Investigator: Cheng Chinhttps://www.ncl.ac.uk/singapore/about/staff/profile/chengchin.html#background Project DescriptionClimate change is an existential threat to humanity. Accurate modelling of the climate is essential for predicting future scenarios of the Earth’s condition such that key policymakers can make informed decisions on mitigation and adaptation strategies. Climate models are extremely complex, combining multiple processes, systems, and cycles into a coupled Earth system model (ESM). Due to limitations in computational capacity with traditional modelling and simulation, current climate models are about 100km is horizontal resolution which is around 2 orders of magnitude coarser than desired, but to reach such fidelity would require around a 10 million times improvement in computational power. Recent efforts have been made in two directions: accelerated computing. The use of heterogeneous hardware such as GPUs , and the use of AI/ML to replace some of the more expensive operations, is important to the project.
|
|
mitoML: Machine Learning to understand mitochondrial disease pathologyPrinciple Investigator: Stephen McGoughhttps://research.ncl.ac.uk/nail/ Project Description"mitoML: Machine learning approaches to understand mitochondrial disease pathology. Read on to find out more. Mitochondrial diseases are currently untreatable, one of the reasons for this is that we don’t completely understand their causes and effects (i.e. pathology). One way of understanding mitochondrial disease pathology is to study protein expressions involved in the functioning of mitochondria. While advancements in protein imaging technologies now allow us to observe many more."
|
|
Chicken RunPrinciple Investigator: Kathryn Garsidehttps://rse.ncldata.dev/kathryn-garside Project DescriptionSmothering or piling behaviour in commercial laying free range broiler chickens is characterised by densely and often fatally crowded groups and has a profound impact on their welfare. Fortunately with some intervention it is possible to interrupt these events and prevent animal losses. This project aims to find methods to automate the identification of such events in videos using machine learning and computer vision methods.
|
|
Structural basis of molecular adaptation to extreme coldPrinciple Investigator:https://www.melnikovlab.com/ Project DescriptionIn this study, we aim to understand the extraordinary capacity of some organisms to survive under extremely hostile conditions. Specifically, we study cold-adapted bacteria that grow and reproduce at temperatures as low as -15°C. We cultivate these organisms in our laboratory and isolate their molecular machines, such as ribosomes. We then study these ribosomes using cryo-EM to determine their atomic structures. In doing so, we aim to provide a better understanding of how biological molecules adapt to extreme environments, such as low temperatures.
|
|
Premature Enhanced Automated Capture of Comfort Knowledge (PEACOCK)Principle Investigator: Arthur HowardProject DescriptionThe proposed research looks at using deep reinforcement learning to facilitate the autonomic navigation of camera-fitted Unmanned Aerial Vehicles deployed to continuously monitor and collect data about the road infrastructure. To realise the goal we need to solve the navigation problem at two levels: global to follow directions from a map of the monitored area, and local to follow fine level details and obstacles of the actual environment. One of the key challenges of the project is to effectively use transfer learning methods such that navigation models built in a simulated environment can be effectively applied and finetuned in real urban or suburban environment with minimal damage to UAVs.
|
|
Intelligent Internet of ThingsPrinciple Investigator: Bo Weihttps://www.ncl.ac.uk/computing/staff/profile/bowei.html Project DescriptionInternet of Things (IoT) devices, such as mobile phones, smartwatches, smart cameras, etc., are ubiquitous nowadays. A huge amount of data are captured from IoT devices for monitoring and context awareness. Although some measurements of surrounding environments can be monitored directly by some sensors (such as temperature, humidity, etc) in IoT devices, many multi-media or multi-modality measurements still need further analysis due to their complexity. Deep learning is a promising technique for data analysis. It achieves good performance in many areas, such as face recognition, natural language processing, etc. In this research, we leverage deep learning to analyse data from IoT devices, aiming at achieving real-time data processing. We also focus on the optimisation of deep learning models to enable them to run on IoT devices and achieve efficient and effective inference.
|
|
Machine Learning for X-ray SpectroscopyPrinciple Investigator: Tom Penfoldweblink Project DescriptionScientific breakthroughs are often strongly associated with technological developments, which enable the measurement of matter to an increased level of detail. A modern revolution is underway in X-ray spectroscopy (XS), driven by the transformative effect of next-generation, high-brilliance light sources e.g., Diamond Light Source and the European X-ray Free Electron Laser, and the emergence of laboratory-based X-ray spectrometers. However, these new kinds of experiments, and their ever-higher resolution and data acquisition rates, have brought acutely into focus a new challenge: How do we efficiently and accurately analyse these data to ensure that valuable quantitative information encoded in each spectrum can be extracted?The objective of this project is to develop and subsequently equip researchers with easy-to-use, computationally inexpensive, and accessible tools for the fast and automated analysis and prediction of XS. We will optimize and deploy deep neural networks (DNNs) capable of providing instantaneous predictions of XS for arbitrary absorption sites, introducing a step change in ease and accuracy of the XS data analysis workflow. Using DNNs, it is possible to reduce the time taken to predict XS data from hours/days to seconds, democratise data analysis, open the door to the development of new high-throughput XS experiments, and allow end users to plan and utilise better their beamtime allocations by facilitating on-the-fly 'real-time' analysis/diagnostics for XS data.
|