IDSAI Funded Pilot Projects - Past Awards
IDSAI Research Awards and Turing-Exeter Pilot Research Grants aim to support early-phase development of ideas for innovative, interdisciplinary data science and AI research across Exeter. Academic members of staff from all Colleges can apply for the time of an IDSAI Research Fellow to work with them on a pilot research project.
You can read about each of the projects from previous rounds of awards below and if you would like to discuss the outcomes of the projects please contact the PIs.
For further information about IDSAI research awards or Exeter-Turing Pilot Grants please contact: email@example.com
Reconsidering the Roman Workshop: examining the processes behind the making of inscribed texts
Collaborators: Charlotte Tupman (Classics and Ancient History / Digital Humanities) and Jacqueline Christmas (Computer Science)
IDSAI Research Fellow: Dmitry Kangin
Description: This pilot project examined on an unprecedented scale the planning processes of ancient Latin inscriptions, one of our major sources of evidence for the Roman world. It developed a machine learning model to locate and extract characters and make a series of measurements to establish the extent to which Roman workshops used specific controlling ‘modules’ in the creation of inscribed texts.
Dr. Dmitry Kangin, IDSAI Research Fellow in Computer Science, worked with Charlotte Tupman and Jacqueline Christmas to apply neural networks to analyse a subset of the almost 40,000 images that have been made available by the Epigraphische Datenbank Heidelberg. As part of his work Dmitry developed a new basis for training Neural ODEs (Ordinary Differential Equations). By combining machine learning and traditional epigraphic methods, the project deepened understanding of Roman drafting and stonecutting processes, and ultimately enhanced the ability to make accurate restorations of fragmentary inscriptions.
IDSAI Research Fellow: Oliver Stoner
Description: Global demand for food is rising and crop production needs to increase in order to provide for a human population expected to be greater than 9 billion by the year 2050. Issues such as water crises, climate change, pests and pathogens can cause significant crop losses and represent major challenges for global food security. In particular, pests and pathogens account for yield losses of up to 20% of the world's harvest each year, with a further 10% loss post-harvest. Our project focusses on fungi, the most important plant pathogens that cause the loss of 125 million tonnes of crops each year. The spread of fungal plant pathogens and the ineffectiveness of available fungicides pose serious threats to global food security.
This project aimed to apply machine learning methods to biological data in order to predict whether fungi will be able to infect different varieties of crops. Being able to accurately predict infection will enable us to forecast food security threats and inform policies that protect crops. Moreover, these methods have the advantage of identifying the key genes, and interactions between genes, that predict infection. These genes therefore, represent potential targets for the next generation of fungicides.
Assessing the risks of harmful algal blooms and impact on marine aqua culture
Collaborators: Dr Theo Economou (Mathematics); Dr Ross Brown (Bioscience), Dr Ian Ashton (Renewable Energy); Dr Fabrizio Costa (Computer Science); Dr Ricardo Torres (Plymouth Marine Laboratory).
IDSAI Research Fellow: Oliver Stoner
Description: Aquaculture is the fastest growing food production sector globally and is vital for future global food security. SW England is a prime area for the expansion of shellfish farming. However, the uptake of biotoxins from naturally occurring Harmful Algal Blooms (HABs) into shellfish is a significant problem. The main risk is intoxication of human consumers, which causes various shellfish poisoning syndromes (from stomach upsets, to amnesia and paralysis).
The vision for this pilot research project was to make a step change in the accuracy of HAB risk prediction both spatially and temporally, thus facilitating operational and development planning for the UK (and wider) mariculture industry. Development of statistical models for predicting HAB risk and impacts on mariculture will have major academic and societal impact. Importantly this project provided pilot data and key learning for the future development of mechanistic models that will be applicable over wider temporal and spatial scales, including future climate change scenarios.
Who Benefits in Policy Trials?
Collaborators: Dr ZhiMin Xiao (Graduate School of Education) and Professor Mihaela van der Schaar (John Humphrey Plummer Professor of Machine Learning, Artificial Intelligence and Medicine, University of Cambridge & Turing Fellow).
IDSAI Research Fellow: Dmitry Kangin
Description: The use of large-scale randomised controlled trials (RCTs) are fast becoming “the gold standard” of testing the causal effects of policy interventions. RCTs are typically evaluated by the statistical, educational, and socioeconomic significance of the average treatment effect (ATE) on the study sample. Interventions that do not have a statistically significant ATE are often discarded as not meaningful and, as a result, usually not implemented more widely, partly due to the failure to recognise the difference between statistical and substantive hypotheses. However, while some interventions may not have an effect on average, they might still have a meaningful effect on a relevant subgroup of individuals, for whom the treatment is beneficial. In some cases, an intervention could thus still provide a net benefit if rolled out. Understanding and identifying for whom a treatment works is therefore critical for policy-makers and society at large.
Recognising the strengths of RCTs as a research and evaluation tool this project proposed an individualised approach to impact estimation. Thanks to recent development in machine learning and the increase in both quantity and quality of research data, it makes greater sense than before to estimate and compare treatment effects at individual and group levels, ultimately generating deeper insights into the causal treatment effects of tested interventions by uncovering what worked, for whom, and by how much. As in any research, the individualised approach, has its limitations. This project looked to validate the individualised approach using both simulations and empirical studies to help practitioners better understand the cutting-edge influence function based approach to the evaluation of machine learning algorithms for ITEs without the need to access counter-factual outcomes.
Multimodel Techniques for the Study of Sexism in Political Institutions
COLLABORATORS: Professor Susan Banducci (QStep, SSIS, Turing Fellow), Prof. Hilde Coffé (University of Bath, Department of Politics, Languages and International Studies), Dr. Tom Fincham Haines (University of Bath, Computer Science).
IDSAI Research Fellow: Dr Ravi Pandit
Description: Sarah Child’s “Good Parliament” Report of 2016, which focused on making the UK House of Commons more inclusive especially for women, first recommendation was that “unprofessional, sexist and exclusionary language and behaviour should have no place in the House”. This sexist behaviour creates a barrier to the full participation of women in policymaking and decreases democratic quality. Therefore, it is crucial to study when, where and how sexism occurs in political institutions. Given that sexism is not one behavioural trait but can be exhibited in many ways (including through voice and gestures), the aim of this project is to explore the feasibility of developing a multimodal technique (relying on text, image and audio) to measure sexist language and behaviour in the House of Commons.
A machine-learning approach to classifying neuron type
COLLABORATORS: Dr Akshay Bhinge (College of Medicine and Health), Dr David Richards (Physics), Professor Krasimira Tsaneva-Atanasova (Mathematics, Turing Fellow), Dr Kate Madden (University of Newcastle).
IDSAI Research Fellow: Dr Ravi Pandit
Description: Motor neuron disease (MND) is a devastating neurodegenerative condition characterized by loss of motor neurons. Understanding the molecular events that lead to motor neuron degeneration in MND is paramount in developing therapies to halt or even reverse the degeneration. Monitoring of motor neuron health in vitro is currently performed using immunostaining where cells are chemically fixed and specific intracellular proteins are stained with fluorescent antibodies. However, immunostaining does not allow real-time monitoring of motor neuron health and can lead to artefacts in neuronal structure. Real-time analysis is important because MND motor neurons display progressive deterioration over extended time periods in vitro. What is critically needed is the ability to identify motor neurons in unstained images. At the moment, however, this is simply not possible. In this project, the team aim to rectify this by developing a novel machine learning approach that can identify neuron type based solely on cell shape without the need for any staining.
AI-enhanced decision support systems for treescapes in the UK
COLLABORATORS: Daniel Williamson (CEMPS, Turing Fellow), Brett Day (Business School), Ian Bateman (Business School), Deyu Ming and Serge Guillas (UCL,Turing)
IDSAI Research Fellow: Dr Bertrand Nortier
Description: The 2019 Climate Change Act amendment requires the UK to achieve net zero greenhouse gas (GHG) emissions by 2050. Crucial to any strategy for meeting this target is the widespread planting of trees, removing GHG from the atmosphere. Design of “treescaping” policies must address two challenges: (i) A variety of factors (such as reliance upon private-sector uptake of incentives) introduce uncertainty regarding the level of GHG removed by tree planting; (ii) Alongside carbon sequestration, tree planting affects a wide array of other ecosystem services including biodiversity, flood, recreation, fire risk, food production and so on.
For each element of (i) and (ii) different research communities have models for their piece of the process. All models are interconnected and can be computationally intensive. To provide policy support, the project aims to run this network of models, propagating their uncertainties over a continuum of policies, in order to present decision makers (DM) with a subspace of policies that are consistent with their stated targets for CO2 sequestration and illustrate trade-offs with costs and other ecosystem services. DM may have an ill-formed prior understanding regarding the trade-offs they might accept across impacts, so rather than using a multi-objective optimisation, the vision is to deliver a mapping of the whole policy space for DMs to explore in real time, offering the potential for radical improvements in policy.
Identifying drug-vaccine interactions in electronic health record data
COLLABORATORS: Prof William Henley (Health Statistics Group, CMH), Prof Mark Kelson (CEMPS; Turing Fellow), Prof Sebastian Vollmer (University of Warwick; Turing Institute), Dr John Dennis (CMH), Dr Lauren Rodgers (CMH), Dr Adam Streeter (Münster Universitätklinikum), Dr Gus Hamilton (NIHR Academic Clinical Fellow, North Bristol NHS Trust).
IDSAI Research Fellow: Dr Oliver Stoner
Description: As progress is made towards developing vaccines against COVID-19, there is increasing interest in understanding ways of enhancing the effectiveness and safety of both new and existing vaccines. Understanding ways of enhancing vaccine response has the potential to have a major impact on tackling morbidity and mortality from existing and emerging infectious diseases in vulnerable populations.
Some of the most common prescription drugs in the UK are medications to treat high blood pressure, statins to treat or delay cardiovascular disease, proton pump inhibitors for heart-burn and acid related disorders, metformin to delay or treat diabetes and non-steroidal anti-inflammatory drugs. A number of these medications are also known to have modulating effects on the immune system that could either inhibit or enhance the immune response when patients receive routine vaccinations. The project team aims to use machine learning approaches to facilitate a data-driven exploration of drug-vaccine interactions to identify candidates for future investigation. This data-intensive screening approach has the potential for rapid deployment at scale to validate interactions with known mechanisms as well as contributing to the discovery of new targets. This proposed project combines methodology development with delivering clinically-relevant results for influenza vaccination.
Transfer learning for clinical applications
COLLABORATORS: Prof David Llewellyn (CMH and Turing Fellow), Dr Neil Oxtoby (UCL), Dr Janice Ranson (CMH), Dr Charlotte James (CMH), Razvan Marinescu, (Postdoctoral Researcher in Medical Image Computing, MIT)
IDSAI Research Fellow: Dr Bertrand Nortier
Description: One barrier to the development and implementation of robust unbiased algorithms is the heterogeneous nature of clinical, population-based and experimental data. Attempts to enhance healthcare have been hampered by the use of single datasets to develop algorithms that are overfitted and don’t perform well in other datasets, contexts or populations. Models developed for rare conditions are particularly problematic due to limited training data and opportunities for external validation. Transfer learning may be a solution to these difficulties, allowing for knowledge learnt from one dataset to be transferred to a second, where the data is similar. The primary objective of the project is to investigate the utility of transfer learning for clinical application. The project will assess whether a machine learning model trained to predict two-year dementia risk can be transferred between clinical samples. As a secondary objective the project will evaluate the utility of Disease Knowledge Transfer (https://arxiv.org/abs/1901.03517) to extend the way in which transfer learning can be used with clinical data.
Automated Detection of Hillforts in South West England
COLLABORATORS: Dr João Fonte (Co-I) and Dr Ioana Oltean (Archaeology), Professor Leif Isaksen (Digital Humanities) Dr Jacqueline Christmas (Computer Science) (Co-I); Dr Albert Chen (Centre for Water Systems); Jane Galwey (Camborne School of Mines)
IDSAI Research Fellow: Dr Dmitry Kangin
Description: This project uses a machine learning approach to test the automated pattern recognition of, and data enhancement for Iron Age hillforts in South West England based on airborne laser scanning data.
Despite the recognition of their great potential the development of automatic detection methods applied to archaeological remote sensing is still in its infancy and limited to very simple morphologies. However, the increasing availability of remote sensing data of higher resolution, higher acquisition frequency and lower cost, demands a paradigm shift, where traditional man-made identification and mapping of archaeological sites cannot be the only option available.
The project will test the extent to which existing algorithms developed for other purposes can assist the detection in airborne LiDAR datasets of hillforts. The project proposes a meaningful solution to protect archaeological heritage and fight the rate of its destruction due primarily to development and climate change globally. If successful, this approach could be expanded and applied to other case studies involving similar archaeological objects elsewhere; other types of archaeological features; potentially other types of datasets (e.g. multispectral aerial and satellite imagery). This will be highly relevant in particular for cultural heritage management by national and local agencies in Britain and beyond.