Cardiovascular diseases (CVDs) is the second biggest killer in the UK and currently, more than 7 million people are living with CVD in the country. Early identification of individuals with significant risk is critical to improve the patient quality of life and reduce the financial burden on the social and healthcare systems. A large number of CVDs lead to the shortage of blood supply to the heart muscle and abnormal motion, which can be diagnosed non-invasively by analysing the patient's dynamic cardiac imaging data. Manual assessment of these images is subjective, non-reproducible, limited to the left ventricle, and time-consuming. Statistical atlases, describing the 'average' pattern of the heart motion over a large healthy population, can be potentially useful to identify deviations from normality in individuals. However, the integration of the existing atlases into clinical practice is inhibited by three key limitations: (i) the derived motion statistics are often independent of the patient's age, gender, weight, etc. (metadata) that are essential for precise diagnosis, (ii) Being non-probabilistic, these atlases fail to provide a measure of certainty in the extracted motion abnormalities thus their clinical reliability is seriously hampered, (iii) they are often derived using a small number of data sets (n<1000), limiting their statistical power. To alleviate these key limitations, this proposal aims, for the first time, to develop a full probabilistic atlas to accurately evaluate bi-ventricular motion abnormalities by holistically integrating imaging and metadata from a large population cardiac imaging study. BIANDA will be a novel Bayesian approach extending the recent developments in deep recurrent neural networks (RNNs). These networks provide a natural mechanism to model sequential data such as 2D video. Yet, using RNNs to model the complex dynamics of the heart motion is conceptually new and evidently powerful. The motion will be modelled as the spatiotemporal (3D+t) sequence of the heart shapes across the full cardiac cycle, extracted from cine Cardiac Magnetic Resonance (CMR) images. The atlas will be a recurrent model that, given a sequence, it will predict a probabilistic distribution function (pdf) for the next status of the heart. More importantly, the pdf will be conditioned on the patient's metadata. Thus by measuring the spatial deviations from the expected shape at each phase, the atlas will allow very accurate quantification of anatomical and functional cardiac abnormalities (and variances showing uncertainties) specific to the patient's age, gender, age, ethnicity, etc. The PI has an extensive experience in developing Bayesian and non-Gaussian statistical atlases from shapes. However, the previous work (i) was not designed to analyse motion data, (ii) discarded the patient metadata (such as age, gender, ethnicity, etc.), and (iii) did not scale into large populations. Therefore, the atlas was not clinically deployable to study cardiac motion abnormalities, which are relevant to various CVDs. This proposal will significantly depart from the PI's previous by combining Bayesian models with deep neural networks. The former is required to handle uncertainties; the latter will significantly boost the prediction and computational efficiency (using GPUs), thus scalability. The atlas will be derived from the UK Biobank CMR study aiming to scan n>100,000 patients by 2022. The training of the atlas will be pursued as the new releases of the data sets from the UK Biobank becomes available. The PI has established collaboration with the clinical advisor for this study and has full access to the CMR data sets. This is essential for the success of the proposal as the training of deep neural networks requires access to an ample of data sets, a possibility which has emerged only recently. In this regard, BIANDA is timely and promising.

We will develop novel, low-cost, diagnostic technologies for the detection of a blinding condition called diabetic retinopathy (DR) in China. These technologies will enable cost-effective, large-scale detection of sight-threatening disease to be performed by non-expert health care workers at the time and place of patient care in China. Over 110 million people in China live with diabetes and this number is expected to increase to 150m by 2040. Diabetes causes visual loss through damage to blood vessels of the retina at the back of the eye. Treatments are effective (and increasingly affordable in China) but only if the disease is detected early enough. Improving early diagnosis and treatment of DR is one of the principal objectives of the Chinese Government's 5-Year National Plan of Eye Health (2016-2020). Current methods of detecting DR rely on costly imaging equipment and many skilled personnel to take and interpret retinal images. China has very few health workers with these skills. Case detection strategies which are cost-effective in Western Europe cannot possibly be replicated at the scale necessary for China. Building on an existing collaboration between the University of Liverpool, Peking University and the Chinese Medical Association (CMA), our joint research team from Liverpool and China of engineers, statisticians, education specialists, eye doctors and health economists are well placed to develop a new diagnostic imaging solution tailored to local needs. Our objectives for this project are: 1. To develop a novel, low-cost, robust imaging device for detection of DR 2. To develop new automated computer algorithms to rate images of the retina for sight-threatening disease 3. To develop a novel comparative judgement method to refine the DR severity grading from automated computer algorithms 4. To validate the new technologies in the UK and test them in China to ensure maximum cost-effectiveness We will develop a new low-cost, camera designed specifically for the needs of China. This device will produce both high-quality colour images and optical coherence tomography (OCT) images of the retina. OCT is widely available but current commercial systems are very expensive. Our new device will be based on our novel patent-pending technology and will be first tested on human donor tissue. We will develop and evaluate new, automated image analysis techniques allowing computers to learn to analyse both colour and OCT images of the retina. We aim to produce systems capable of differentiating between patients with and without DR and between those with mild/moderate and severe disease at the time and place of patient care. In order to achieve a high level of diagnostic accuracy (over and above automated image analysis), we will develop and evaluate a new human learning system. This system will harness the collective judgements of Chinese health workers to rank images in terms of DR severity. We will create a self-sustaining network of activity where novices and experts support each other in making effective clinical judgements. We will develop new statistical methods to underpin the system and to evaluate both diagnostic accuracy and performance of the health workers. Our imaging and diagnostic system will be validated in 241 patients with diabetes in the UK. Our project team in China will then undertake a pilot study of 461 patients and a costing exercise. The diagnostic accuracy of the system and its cost-effectiveness will be investigated. We will engage policy makers through our Chinese team who occupy leading roles in the CMA. Detection and treatment of sight threatening DR will prevent disability with benefit to Chinese society and China's economy. Our systems will upskill health care workers, strengthen existing health systems and build research capacity in China. Dissemination of our techniques through open-source software will maximise benefits for other low and middle-income countries.

This proposal brings together 19 universities, including 12 out of 16 newly established UKRI CDTs in Artificial Intelligence. Led by the University of Oxford, with support from the Alan Turing Institute (Turing), Bath, Bristol, Cambridge, Exeter, Imperial, KCL, Leeds, Loughborough, Newcastle, QMUL, Sheffield, Southampton, Surrey, Sussex, UCL, Warwick and York, our proposal aims to build on the success of the JADE Tier 2 facility. The current JADE facility represents a unique national resource providing state of the art GPU computing facilities to world leading experts in the areas of Artificial Intelligence/Machine Learning (AI/ML) and molecular dynamics (MD) research. In addition to providing a leading compute resource, the JADE facility has also provided a nucleus around which a national consortium of AI researchers has formed, making it the de facto national compute facility for AI research. By providing a much-needed shared resource to these communities, JADE has also delivered an outstanding level of world leading science, evidenced in the twenty two pages of preliminary case studies submitted to EPSRC on 11/09/18. JADE2 will build upon these successes by providing increased computational capabilities to these communities and delivering a stronger, more robust service to address the lessons learned from the initial service. The architecture for JADE2 will be a similar to that of JADE, based on NVIDIA's DGX platform. JADE is formed from 22x DGX1V nodes. JADE2 will be over twice the size of JADE and employ the more cost effective DGX1 Max Q platform. Differences between Max Q and the premium DGX1V are centred on on a slightly reduced bandwidth to GPU memory and lower peak compute performance. Tests of relevant codes on these platforms show that, for AI/ML and Molecular Dynamics, Max Q achieves at least 3/4 performance, using 2/3 the power for 1/2 of the price. The system will be run as a national facility, providing free access to all academic users through a lightweight Resource Allocation Panel (RAP). HECBioSim will run the RAP for MD users, ATI will run the corresponding RAP for AI/ML users.

We live in an era of abundant data. Rapid technological advances, such as the internet, have made it possible to collect, store and share large amounts of information more easily than ever before. The availability of large amounts of data has had a major impact on society, commerce, and the sciences. Data plays a particularly important role in the sciences. Data is what you get from conducting experiments, and data is what you use to test scientific theories. In recent years, the amount of data collected and generated in the sciences has grown tremendously. We need better tools to model this data, so that we can understand and test theories and make scientific predictions. Our proposal focuses on advanced statistical tools for modelling data. It is important that the models are based on probability and statistics, because any model of real world phenomena has to represent the uncertainty we have from incomplete information and noisy measurements. Probability theory provides a coherent mathematical language for expressing uncertainty in models. Our proposal develops models based on Bayesian statistics, which used to be called ``inverse probability'' until the 20th century, and refers to the application of probability theory to learn unknown quantities from observable data. Bayesian statistics can also be used to compare multiple models (i.e. hypotheses) given the data, and thus can play a fundamental role in scientific hypothesis testing. We will develop new computational tools for Bayesian modelling, ensuring that the models are flexible enough to capture the complexity of real-world phenomena and scalable enough to deal with very large data sets. We will also develop new methods for deciding which data to collect and which experiments to perform, which can greatly reduce the cost of scientific inquiry. We will make use of the latest advances in computer hardware, in the form of massively parallel graphics processing units (GPUs) to speed up modelling of scientific data. This proposal is truly cross-disciplinary in that we do not focus on a single scientific discipline. In fact, we have assembled a team whose expertise spans Bayesian modelling across the physical, biological and social sciences. We will create modelling tools for better astronomical surveying of the skies so that we can understand the composition of our universe; we will create tools for analysing gene and protein data to so that we can better understand biological phenomena and design drug therapies; and we will develop powerful methods for modelling and predicting economic and financial data which will hopefully reduce risk in financial markets. Surprisingly, these diverse areas of the sciences---astronomy, biology and economics---can come together through a unified set of computational and statistical modelling tools. Our advances will benefit not just these areas but many other areas of science based on data-intensive modelling.

This proposal is to fund the Isambard 3 Tier-2 national HPC service from the GW4 Alliance universities of Bristol, Bath, Cardiff and Exeter. Isambard 3 will follow on from the successful Isambard 1 & 2 projects, delivering a supercomputer service based on the increasingly important Arm architecture. Isambard 3 will employ NVIDIA's new 'Grace' Arm-based processors, due out in early 2023, to build a fast yet highly energy efficient service. Isambard 3 will be moving into its own dedicated facility, hosted by the University of Bristol. This new facility, homed within a new modular data centre, will for the first time employ direct liquid cooling to the processors, reducing the total amount of energy used for cooling the system, and creating the opportunity to extract and use the waste heat. We expect to procure a system of at least 6 racks, 55,000 cores, which should be large enough to rank in the Top500 list of supercomputers at SC in Nov 2023. Isambard 3 will also continue to provide a Multi Architecture Comparison System, or MACS, which will includes 2-4 nodes of every important new CPU and GPU architecture to be released during the 4 year lifetime of the Isambard 3 service. This will enable scientifically rigorous architectural comparisons between Arm and the latest CPUs and GPUs from Intel, AMD, NVIDIA, and other emerging providers. The Isambard 3 service will feature an expanded support team compared to previous services, including 6 RSEs and 2 systems administrators on the technical support side, as well as 0.5 FTE of project manager, and a business development manager on the user community engagement and development side. In summary, the successful Isambard Arm-based Tier-2 service will be expanding to support a much wider user community across all of UKRI, with the first Arm-based server CPUs optimised specifically for HPC from a top-tier silicon vendor. It will be one of the first such systems anywhere in the world, and deliver one of the most energy efficient, low carbon, general purpose CPU services available. The Grace processors should be amongst the fastest available for Isambard's user community, which we expect to expand into weather and climate applications, as well as life sciences and beyond.