Aiming to manage air pollution reduction. How can other data measurements from Satellites, Sensors and surveys or other sources be used to determine where emission occur and why air pollution is so high in urban zones in the UK and India.

.

What methods can we use to model power dynamics and responsibility hierarchies in AI systems to improve crisis response at the executive level?

In high-stress crisis scenarios, executives often face delays in decision-making due to unclear role definitions, overlapping responsibilities, and power imbalances. These challenges are exacerbated by crisis management plans that often “sit on a shelf” unread until a crisis occurs. Without active understanding or use, such plans fail to prepare leaders for high-stakes situations. We aim to develop an AI-driven tool that dynamically clarifies roles, mitigates power differentials, and provides actionable insights to streamline decision-making in real time. This tool should ensure that crisis management frameworks are actively integrated into preparation and response processes, rather than being left unused until an emergency.

How can we train AI systems to prioritize ethical decision-making and compliance with international standards during crisis management?

Crisis management often involves navigating ethical and regulatory complexities, especially in industries like mining, insurance, and defence. While crisis management plans may reference these standards, their lack of integration into real-time decision-making processes leaves organizations vulnerable to ethical oversights. The CoordinAIt platform seeks to embed ethical frameworks directly into its AI models, ensuring compliance with international standards and prioritizing human and environmental considerations. This approach would address the shortcomings of static, underused crisis plans by providing dynamic, ethics-driven recommendations that guide organizations during high-pressure situations.

How can we use technology to automate analysis of complex mixtures of chemicals (for example body odour)?

Taking inspiration from insects, Arctech Innovation uses odour to solve problems in medical diagnostics and pest control. We produce data on odours mixtures, for example we can characterise the individual chemical components and their concentrations within the smell of human skin. When you are ill, your smell changes, and we can use this change to diagnose infections and other conditions. In previous work we have manually compared odour profiles of healthy and unhealthy individuals to look for these characteristics odour changes. However, as we grow and scale up we are looking for a way to automate and improve this analysis step. The data comes from analytical chemistry. Gas chromatography separates mixtures by weight, and we can use different methods to identify the compounds present: retention time, area under the curve. Differences between groups could be a change in a single chemical, but more commonly is a change in relative ratios of a number of key volatile compounds.

How can we choose the best connectivity methods for different IoT products?

We are building smart products for a range of applications:
• detecting insect or other pests in houses and hotels
• detecting insect or other pests in storage or transport settings
• detecting animals inside burrows
• detecting ill health in animals in zoos or on farms
• detecting ill health in animals and humans in veterinary or clinical settings
• detecting ill health in humans in non-clinical settings (e.g. border control)
This means we have a plethora of battery, range and data considerations when thinking about connectivity options. We have spoken to some companies offering solutions, but they often provide contradictory advice. What are the key factors to consider, and is any one method likely to be a best fit?

How might we develop an open source digital platform that allows a local mutual credit system to be tested and applied among Edinburgh’s social enterprise and local small business community to contribute towards a circular economy?

This project seeks to develop a digital mutual credit system that
incentivises Circular Economy practices and generates data-driven insights to guide research, investment and policy. Our development of a complementary economic system that changes how and what we value is a response to the twinned crises of ecological degradation and economic pressures faced by local businesses.

How do you suggest we nudge the cycle-based sparse graph optimization to achieve faster convergence and accurate labelling of multi-million edges real graph?

.

With the high dimension vector retrieval scenario, especially more and more application as KV cache compression ,is there any mathematics principle behind to accelerate the retrieval based on information theory or High-dimensional statistics?

.

How should a worldwide corporate real estate organisation monitor biodiversity to maximise its impact on biodiversity at its sites and surrounding areas, and be best able to respond to new and emerging biodiversity standards such as CSRD and the European Sustainability Reporting Standard?

IBM Global Real Estate operates in 95 countries and has 650+ sites. Approximately 90% of the sites are offices, and IBM also has manufacturing, research, and data centre sites. Nine sites are on or within 2.5km of biodiversity-sensitive protected areas.

IBM has technology that could be used, such as the Environmental Intelligence Suite (used for environmental risk) and Maximo Monitor, which can capture images from drones, robots, and IOT devices. IBM would be open to discuss collaborations for student projects using IBM technology for biodiversity-related topics. IBM is also open to discuss both technical and non-technical collaborations using our sites with landscaping activities for case studies that advance biodiversity studies.

How can AI be used to enhance fraud detection systems and ensure cybersecurity in highly regulated sectors?

Legal-Pythia seeks to improve its AI-driven fraud detection tools, and we also aim to develop AI-powered cybersecurity tools that align with strict data privacy regulations, specifically in sectors that are highly regulated, such as finance and transportation. Insights into the best AI methodologies for identifying fraudulent activities would help refine our solutions, ensuring higher accuracy and compliance with legal frameworks, while understanding how to balance cybersecurity protection with privacy requirements will enable us to create solutions that are both effective and compliant with laws such as GDPR.

Contested Logistics Operations– use of AI capabilities to support with resilience and operations continuity in challenging logistics operations environments– inc. forecasting and stock management/planning (AUKUS focus). Including Digital Twin for Large-scale supply chain scenarios (critical civil, government and defense sectors) – including modelling, logistics chain modelling, warehouse optimization.

To expand this challenges of “but when one of those “nodes” go down, how do we recover” – It IS about generating neural network pathways on the fly when those nodes go down, based on the current landscape, but also thinking about how we can do things differently – a static warehouse is an easy target for adversaries, so can we put a load of stock in trucks and drive them around so it’s harder to take out our stores? Can AI then work out the loading of those trucks to ensure we distribute (inside the trucks) the right stock in the right areas for quick fulfilment of an order. Can we look across our allies’ positions to see where we might be able to go and ask a friendly nation for stock that we can no longer get ourselves?

Edge, Adaptive and Sustainable AI – deployment of AI algorithms and AI models directly on local edge devices such as sensors or Internet of Things (IoT) devices, which enables real-time data processing and analysis without constant reliance on cloud infrastructure. Focus on power consumption optimisation, optimisation of ML algorithms to improve performance in terms of power consumption (not just hardware but research into ways to rewrite certain ML primitives to be more efficient in training and/or inference).

How does the alternative hardware design that Dr. Serb presented work with common compression techniques. What benefits does alterative hardware offer in processing power, accuracy, and customization that common products like NVIDIA Jetson do not? FPGA typically suffer from code sustainability – i.e., having to update the code is much more cumbersome which will likely happen with AI since we should be maintaining it over its lifecycle.

What training and adaptation techniques are necessary to enable LLMs to provide culturally sensitive and linguistically appropriate communication across diverse populations, particularly given imbalanced representation in both data and within the population at large?

Cultural nuances are critical in effective communication. Exploring approaches to minimise bias and maximise relevance is of particular interest.

We wish to explore the accuracy and effectiveness of using Artificial Intelligence in historic aerial photography to extract and identify features and automatically georeference images using modern-day geospatial data.

To view many examples of the historic aerial photography in question please visit: https://ncap.org.uk/ https://www.ncap.org/

What would be the single best indicator to choose to monitor the health of our local economy?

Local and national governments host vast amounts of data across various areas of civic life. How can we effectively monitor local improvements across so many interconnected domains? Additionally, how can we communicate this information in a way that instils confidence in the metrics we use?

With the current National Government narrative focusing on mission-based government across three foundations, five missions, and six milestones, it can be confusing for those who are following along. How can local governments utilise better metrics to communicate a sense of optimism within their communities?

Key indicators for growth in a local economy may include:
• Employment Rates: High employment rates indicate a healthy economy with job opportunities for residents
• Gross Domestic Product (GDP): Measures the total value of goods and services produced within the local economy + (per capita)
• Personal Income Levels: Rising personal income levels suggest improved economic well-being and spending power of residents
• Business Activity: The number of new businesses, business expansions, and overall business activity reflect economic vitality
• Real Estate Activity: Trends in property values, construction, and real estate transactions can indicate economic growth
• Industry Concentrations: The presence and growth of key industries can drive local economic development
• Consumer Spending: Higher consumer spending levels can stimulate local businesses and services
• Infrastructure Development: Investments in infrastructure such as transportation, utilities, and technology can support economic growth

How do we create synthetic individual level health datasets which have the same properties as the source data, using machine learning and/or AI, and which meets data disclosure requirements of the system.

As a country we want to fully engage with the potential of AI and particularly within health and social care. With methodologies for creating synthetic datasets, we will be able to support many more collaborations around health data and the use of AI. They could help to identify groups of people that need support, help to influence policies, and support change in the health and social care system.
The datasets could cover services, demand and capacity, enabling us to look at long term and emerging trends (e.g. climate change, demographic, disease prevalence) and service redesign (e.g. emerging technologies and workforce constraints), to support the transformative change it will require to be sustainable going forwards. The health & social care system is under major pressure and intelligence to support its transformation will be vital.

How can AI help us improve workload predictions and proactively identify resource issues before they cause delays or blockages in our projects? Specifically, we’re interested in AI strategies that can enhance our ability to predict future demands and monitor trends to alert resource planners to potential problems early

Retain Cloud offers advanced resource management software designed to help customers maximize the utilization of their resources. Currently, resources are manually added to meet the role requirements of each project and booking. To further enhance efficiency and precision, we seek to leverage AI and ML to address two of our main resource management challenges for our customers. Our current resource management process faces significant challenges: 1. Accurate Workload Prediction: The existing system struggles to predict future workload demands accurately, leading to inefficient resource allocation, missed project deadlines, and increased costs. To resolve this, we plan to implement a machine learning-based workload forecasting model. This model will provide reliable predictions, enabling better planning and decision-making. 2. Proactive Impact Assessment: Our process lacks the ability to proactively assess the impact of changes in resource allocation, project timelines, and external factors. This results in inefficiencies and heightened risks. We aim to integrate scenario simulation tools into Retain Cloud’s resource management platform. These tools will empower clients to plan for multiple scenarios and mitigate risks effectively. Resource planners review the available resources against a number of different criteria, key skills, soft skills, wider knowledge , experience with the client and certifications The weighting used by the resource planners will change from customer to customer and role to role.

Resource utilisation: optimising for people not just numbers:
https://www.retaininternational.com/blog/resource-utilisation-optimising-people-not-just-numbers

Understanding Resource Capacity Planning:
https://www.retaininternational.com/blog/understanding-resource-capacity-planning

AI-Powered Tools & Software For Resource Management:
https://www.retaininternational.com/blog/ai-powered-tools-and-software-resource-management

How can AI be used to improve the daily operations of resource planners, helping them make more accurate decisions, optimize resource allocation, and proactively address potential issues before they impact project timelines or costs?

Retain Cloud offers advanced resource management software designed to help customers maximize the utilization of their resources. Currently, resources are manually added to meet the role requirements of each project and booking. To further enhance efficiency and precision, we seek to leverage AI and ML to address two of our main resource management challenges for our customers. Our current resource management process faces significant challenges: 1. Accurate Workload Prediction: The existing system struggles to predict future workload demands accurately, leading to inefficient resource allocation, missed project deadlines, and increased costs. To resolve this, we plan to implement a machine learning-based workload forecasting model. This model will provide reliable predictions, enabling better planning and decision-making. 2. Proactive Impact Assessment: Our process lacks the ability to proactively assess the impact of changes in resource allocation, project timelines, and external factors. This results in inefficiencies and heightened risks. We aim to integrate scenario simulation tools into Retain Cloud’s resource management platform. These tools will empower clients to plan for multiple scenarios and mitigate risks effectively. Resource planners review the available resources against a number of different criteria, key skills, soft skills, wider knowledge , experience with the client and certifications The weighting used by the resource planners will change from customer to customer and role to role.

Resource utilisation: optimising for people not just numbers:
https://www.retaininternational.com/blog/resource-utilisation-optimising-people-not-just-numbers

Understanding Resource Capacity Planning:
https://www.retaininternational.com/blog/understanding-resource-capacity-planning

AI-Powered Tools & Software For Resource Management:
https://www.retaininternational.com/blog/ai-powered-tools-and-software-resource-management

What are the feasibility, effectiveness, and potential challenges of implementing VR-based rest and recovery spaces for paramedics, designed to reduce stress, alleviate fatigue, and improve readiness for subsequent emergencies?
Given that paramedics are often unable to leave patients at hospital for extended periods due to hospital pressures, frequently missing their allocated breaks, how could such an intervention be optimised to integrate seamlessly into the high-pressure workflows of urgent and emergency healthcare systems? Additionally, what measures could best evaluate its impact on paramedic well-being, morale, and service availability?

Paramedics often work long and demanding 12-hour shifts, during which they are entitled to allocated rest breaks to recharge and maintain their well-being. However, due to significant operational pressures, these breaks are frequently missed. A common scenario involves ambulances queuing at hospitals, where paramedics are required to remain with patients for extended periods until handover is complete. This prolonged engagement not only prevents them from taking formal rest breaks but also compounds the physical and mental strain of their roles.
The impact of missed rest breaks on paramedic crews is profound. It directly affects morale, with staff reporting increased stress and dissatisfaction, and raises the risk of fatigue-related issues. Fatigued crews are more likely to report being unfit for work, leading to reduced availability for subsequent emergencies. This creates a cascading effect, increasing pressure on the ambulance service and potentially compromising patient safety. Recognising these challenges, the Scottish Ambulance Service (SAS) is committed to supporting frontline staff by exploring simple yet effective measures to boost morale and well-being.
Whilst there is strong evidence that providing rest and relaxation spaces can significantly reduce short-term perceived stress among healthcare workers, this initiative represents a novel approach. We believe this will be the first time such a question is being explored on a national scale across Scotland, focusing specifically on the use of VR technology and non-fixed, mobile spaces. By evaluating the potential of these innovative interventions, the SAS aims to not only enhance paramedic well-being but also improve service availability and ensure better outcomes for patients.

Can we train an existing language model (e.g. Amazon Comprehend) to identify PPI and environmentally sensitive information and automate the processing of redactions.

We have performed project scoping and user needs with support from SG Data Science team. We have datasets available. Huge opportunity for cost savings and use of LLMs and automate to perform tasks better than human operators.

How can we predict the impact energy efficiency measures on house value?

Decarbonising homes is critical to the UK hitting its net zero commitments. Unfortunately, many measures are uneconomic and payback periods are often measured in decades. Reflecting the impact of retrofit on the value of a house can help make the business case more attractive. Existing data sets (EPCs, sold prices etc) can be used to build a model to estimate this increase

How can we calculate health transport spend at a multi-health board, regional level?

We are the regional transport partnership for the South East Scotland, and are working with our four NHS health boards; Lothian, Forth Valley, Borders and Fife, to consider regional responses to improving transport to health. We are aware that there is significant public spending across health boards, local authorities, and Scottish Government to provide transport to health, including a significant but complex taxi spend.

We would be interest in your thoughts on a methodology to calculate this spend, particularly when some of it is not publicly available information and different organisations account for this spend in different ways.

Use of supervised learning requires generation of datasets with high quality labelling. When developing products, such labelled datasets are difficult if not impossible to define. What techniques exist to address this challenge, and can GenAI play a role to support organizations who wish to take advantage of machine learning early in the development process?

At ST in Edinburgh we have lots of cases where we begin with small datasets, imbalanced and variable – and the labelling process is either too costly to perform or the early stage data is too variable. However, the early stage development process provides indicative data which may be useful as we begin to enter production ramp. We are keen to understand what techniques exist to help us make maximum use of the data we have available to develop supervised machine learning models as early as possible.

How do we address concerns around sensitive data leakage when training and sharing machine learning models across organizational boundaries spanning academia, industry and clinical/healthcare settings, especially in the context of pre-trained foundation models?

Collaborative projects often lead to models trained on data accessible to only a subset of partners (e.g individual level patient data held by clinical organizations or proprietary chemical/biomolecular libraries from industry). There are legitimate concerns in sharing trained models due to possibility of data leakage, leading to an inability to carry on further research and development. How do we mitigate these risks while still ensuring effective translation of such research?

How might AI be utilised to assess and improve the sustainability of projects financed by UKEF?

UKEF ensures its commitment to sustainable financing by integrating environmental, social, and governance (ESG) criteria into its decision-making processes. UKEF is looking to support more projects that align with global sustainability goals. Generative AI could help in analysing project proposals, historical data, and environmental reports to ensure that financed projects meet stringent ESG standards using a less manual approach to researching, compiling and evaluating the environmental and social impact of a project. In addition to how AI could be used to reduce the time spent on the assessment activities, what data sources and AI techniques are most effective in evaluating ESG criteria and predicting the long-term environmental impact of financed projects?

How might AI-driven tools assist in identifying export opportunities, providing tailored advice, and streamlining the application process for export financing including providing personalised marketing to potential SME customers?

UKEF seeks to further expand its support for small and medium-sized enterprises (SMEs) to help them access international markets and grow their export capabilities. SMEs often face challenges in navigating export markets and accessing financing. Generative AI could analyse market trends, identify potential export opportunities, and provide personalised recommendations and marketing to SMEs. Additionally, AI could streamline the application process by automating document review and risk assessment. Necessary data could include SME profiles, market analysis reports, and historical financing data.

There is a motivation to mitigate impact of offshore wind on the aviation sector by numerically modelling wind turbine blade radar cross-section (RCS) to provide a training set for a machine learning algorithm. However, the modelling of RCS is computationally expensive. What techniques could be used to make the modelling approach viable?

.

The cumulative noise from mooring lines used for floating wind turbines may adversely affect the environment. How could acoustic recording of underwater noise be used to train a machine learning algorithm and relate noise levels to environmental data and allow predictions of future cumulative noise from floating wind farms in the North Sea?

.