Aim
The Strategic Defence Review 2025 – Making Britain Safer: secure at home, strong abroad, highlights how rapid and unpredictable technological advances are changing the character of warfare, as well as the importance of digital enablement in ensuring an integrated, connected, and agile combat force.
Reservoir Computing (RC), a cutting-edge type of in-memory computing, may prove to be increasingly important in many emerging technology areas such as intelligent robotics, artificial intelligence (AI) and digital twins for science, 6G networks, power-efficient data centres, and the Internet of Things (IoT), particularly where fast prediction and computation, or dynamic control is required.
There is a pressing need, therefore, to develop an improved understanding of the capabilities and integrability of RC, as part of Defence’s future hybrid computing system.
Consequently, UK Research and Innovation (UKRI) and EPSRC, in collaboration with Defence Science and Technology Lab and UK government partners, are inviting applicants to attend a joint sandpit to deliver new, innovative, multidisciplinary and transformative approaches to RC for national security and defence.
A collaborative sandpit approach has been chosen to generate research applications that:
- take into account the needs of UK defence and security stakeholders from across government
- form new collaborations between researchers, innovators and government users of research (stakeholders) in diverse research areas
- create new and transformative research ideas in RC, and allowing researchers to pitch projects for funding to test and de-risk novel ideas
- address key research challenges that are identified and described at the sandpit
- can be led by researchers who have not worked in this sector before
We encourage applications from individuals who have had no prior involvement with the defence or security sector, and we welcome applications from those with existing links. It is our intention that participants at the sandpit will remain engaged with stakeholders from the defence and security sectors and be inspired to form longer term collaborations.
Sandpit
The sandpit will be an intensive, inclusive, interactive and creative environment, supporting a diverse group of participants from a range of disciplines and backgrounds in UKRI’s remit to work together.
We recognise the value in enabling collaboration across disciplines which may not usually come together to address the challenges being tackled. The unique opportunity provided by this sandpit will be that attendees will have access to government stakeholders, to drive the research towards real-world scenarios.
The sandpit will be overseen by a director, who will be supported by a team of mentors. The director, mentors and stakeholders will attend the sandpit but will not be eligible to receive research funding. Instead, their role will be to assist participants in defining and exploring challenges in this area. The director and mentors will act as independent reviewers, making a funding recommendation on the emergent projects.
The sandpit process can be broken down into several stages, these are:
- defining the scope of any research to address the UK’s defence and security challenges
- cultivating a common language and terminologies amongst people from a diverse range of backgrounds and disciplines
- sharing understandings of the challenges, and the expertise brought by the participants to the sandpit, and perspectives from relevant stakeholders
- immersing participants in collaborative thinking processes and ideas sharing to construct innovative approaches
- capturing the outputs in the form of highly innovative research projects
- a funding decision on those projects at the sandpit using ‘real-time’ peer review
Scope
Reservoir Computing (RC) is a non-traditional computing architecture that utilises a high-dimensional, nonlinear reservoir to perform computational tasks and implement Machine Learning (ML).
The nonlinearity of the reservoir, which in effect creates memory of previous inputs, means it is well suited to processing observed time-series data gathered from dynamic systems (and it is not optimised for non-time-series tasks): thus, RC has the potential to be important in areas where dynamic control of systems is required, for example intelligent robotics and cyber-physical systems.
Training RC systems requires less computational time and data than more traditional ML architectures. Also, reservoirs can be realised in various media (physical, electronic, biological, quantum etc).
However, the high dimensionality of the reservoir and current training methods mean the architecture is not fully optimised, thus it may not reach the state-of-the-art of very large and optimised deep learning models. RC appears most relevant to use cases in which model adaption (due to low need for training data), low computational burden and dynamic control are required; co-design of algorithms with the appropriate substrates may enable computation in challenging physical environments. These characteristics tend to suggest RC will have most utility at the edge of networks and on small, disconnected devices.
Purpose of research
An understanding of what combination of RC algorithms, architectures and substrates works best for what Defence tasks would be a groundbreaking tool in Defence’s future compute toolbelt, where the objective is to deploy paradigms where they offer the greatest utility. In other words, the purpose is to develop an improved understanding of the capabilities and integrability of RC, as part of Defence’s future hybrid computing system. This may be achieved in several ways, including demonstrating the ability to co-design algorithms, architectures and substrates to meet a specific Defence use case. An example would be in an extreme environment, where low power adaptable compute is required to enable the autonomous behaviour of attritable and consumable assets.
Participants at the sandpit will be introduced to a number of defence and security scenarios by users of technology from across government and will be encouraged to approach problems in an interdisciplinary manner. For that reason, we encourage applications from a range of disciplines from across UKRI’s remit including but not limited to:
- computer sciences
- AI and Machine Learning
- computational biology
- computational neuroscience
- materials science
- physics
- robotics
- mathematical sciences
- signal processing
- engineering
- high performance computing
- modelling and simulation
- digital twinning
- psychology
- responsible research and innovation
Duration
The duration of any successful award is a maximum of 24 months.
Funding available
The FEC of your sandpit project can be up to £1.37 million.
EPSRC will fund 80% of the FEC.
Accommodation will be provided during the residential component of the sandpit. However, participants must make their own travel arrangements. Travel and subsistence costs will be reimbursed.
Since this sandpit is partially residential, and where employers cannot help, EPSRC, in line with UKRI policy, will cover the costs of any additional childcare or caring responsibilities, which is deemed necessary during this period.
Trusted Research and Innovation (TR&I)
UKRI is committed in ensuring that effective international collaboration in research and innovation takes place with integrity and within strong ethical frameworks. Trusted Research and Innovation (TR&I) is a UKRI work programme designed to help protect all those working in our thriving and collaborative international sector by enabling partnerships to be as open as possible, and as secure as necessary. Our TR&I Principles set out UKRI’s expectations of organisations funded by UKRI in relation to due diligence for international collaboration.
As such, applicants for UKRI funding may be asked to demonstrate how their proposed projects will comply with our approach and expectation towards TR&I, identifying potential risks and the relevant controls you will put in place to help proportionately reduce these risks.
See further guidance and information about TR&I, including where applicants can find additional support.
