Efficient, structured and controllable AI systems

Overview

Innovate UK is investing up to £1.66 million to support UK SMEs developing and experimentally validating novel, reusable artificial intelligence (AI) mechanisms. The competition aims to establish the foundations for future AI and machine learning (ML) systems that are more efficient, adaptive, structured and controllable.

The funding is aimed primarily at research-intensive AI startups, deep tech SMEs, university spinouts and pre-commercial businesses developing proprietary AI technologies.

Projects should explore fundamental advances in learning mechanisms, adaptive architectures, causal modelling, relational computation, compositional representations or measurable agent control.

Projects must begin at Technology Readiness Level (TRL) 1 to 3 and are expected to advance by 0.5 to 1 TRL during delivery.

Successful projects must deliver a prototype algorithm, architecture or subsystem; an experimental validation environment; results against defined baselines; ablation and negative control evidence; and a plan for further development.

A technical white paper must also document the project's outcomes, validation evidence, outstanding technical risks and innovation defensibility.

This is phase 1 of a potential two-phase competition. Any future phase is subject to approval and has no confirmed funding allocation.

Scope

Projects must advance a new core AI or ML technology by establishing experimental evidence for a novel architecture, learning process, representation or control mechanism.

The technology must have the potential to support capabilities, products, platforms or services across multiple future applications or markets.

Applications must demonstrate a clear route to defensibility, such as protectable intellectual property, proprietary data advantages, specialist know-how or other credible barriers to entry, alongside a route to scale.

Projects are not required to develop a complete commercial product. However, they must demonstrate the technical feasibility of the core mechanism, architecture or critical subsystem.

Projects must start at Technology Readiness Level (TRL) 1 to 3 at the point of application and are expected to advance by 0.5 to 1 TRL during delivery.

Proposals must demonstrate that AI or ML innovation is the core technical contribution and the principal source of potential competitive advantage.

Projects must deliver:

• Experimental validation of the proposed hypothesis
• A clear validation methodology against predefined metrics
• Evidence of technical novelty and a technical asset
• A scaling rationale for the technology and business model
• A technical white paper

To be in scope, proposals must sufficiently describe:

• The technical novelty and hypothesis-driven experimentation statement
• The technical asset that the project will create
• How the technology could underpin future products, platforms or services
• The justification for the project's TRL classification

Key themes and topics

Proposals must fall within one or more of the following four themes.

Projects must focus on at least one specific priority within each selected theme. Applications that do not align with their selected theme and specific priority will not be sent for assessment.

Theme 1. Resource and data adaptive learning

Learning mechanisms that reduce dependence on large datasets, repeated full model training or unnecessary computation by adapting what information is acquired, retained or updated according to uncertainty, task requirements or expected information value.

Projects must focus on one of the following priorities:

• Active and probabilistic learning, including uncertainty aware selection of informative data, experiments, simulations or interactions
• Continual and selective learning mechanisms that enable new knowledge or capabilities to be acquired without repeated full model retraining
• Data efficient learning under limited or expensive observations, using probabilistic representations, structured priors or uncertainty to improve learning efficiency

Theme 2. Causal and intervention grounded models

Mechanisms that learn, discover or exploit causal and counterfactual structure to improve generalisation, prediction, intervention selection or decision making when underlying conditions or data generating processes change.

Projects must focus on one of the following priorities:

• Causal representation learning
• Interventional or counterfactual learning
• Causal discovery under partial observability

Theme 3. Learned modular computation and routing

Learning mechanisms that discover, select, compose or reuse internal computational components, enabling models to dynamically route computation and reuse learned functions across tasks, environments or novel combinations.

Projects must focus on one of the following priorities:

• Learned modular architectures and computational routing
• Reusable learned functional primitives or subroutines
• Adaptive composition and reuse of learned computational components

Theme 4. Geometric, equivariant and physics structured learning

Learning mechanisms that exploit or discover geometric, symmetry, relational or physical structure to improve data efficiency, generalisation, robustness or consistency beyond what can be achieved by learning these regularities from data alone.

Projects must focus on one of the following priorities:

• Equivariant or invariant learning architectures
• Geometric and relational learning on structured or non-Euclidean domains
• Physics structured learning, including methods incorporating physical symmetries, constraints, conservation structure or differentiable dynamics

Projects that will not be funded

Innovate UK will not fund projects that:

• Do not sufficiently provide clear background IP ownership or rights
• Do not sufficiently provide a specific defensibility route
• Do not align with the competition theme and specific priority areas
• Do not sufficiently provide a named baseline with metrics and numeric targets for validation
• Do not provide a falsifiable hypothesis-driven experimentation plan
• Do not sufficiently address the scope of the competition
• Are primarily literature review studies or requirements gathering without substantive experimental research and development
• Propose routine integration, deployment, orchestration or productisation of existing AI tools or third-party models without novel AI or ML development
• Are not delivering measurable and specific objectives
• Are primarily routine integration or deployment of existing AI tools without substantive technical innovation
• Do not have clear technical novelty and feasibility challenges
• Do not result in defensible foreground IP
• Focus primarily on non-AI or non-ML research and development, including cross-cutting projects
• Focus on prompt engineering, generic agent orchestration, routine Graph Neural Network (GNN) application or passive-only agentic layers

Projects also cannot be funded if subsidies are conditional on export performance or the use of a specified percentage of domestic inputs.

Project duration

1 to 3 months

Must start on 1 April 2027

End by 30 June 2027

Award value

Total eligible project costs between £50,000 and £150,000.

Grant funding is calculated as a percentage of eligible project costs, depending on organisation size and research category. The maximum published funding rate is 70%.

Funding rates

For feasibility studies, which evaluate a project's potential by uncovering key factors and informing confident, well-founded decisions about moving forward, funding is available for eligible project costs of:

• Up to 70% for a micro or small organisation
• Up to 60% for a medium-sized organisation
• Up to 50% for a large organisation

For industrial research projects, purposeful research that builds new knowledge and skills to improve or develop products, processes or services, often through prototypes or system components that validate ideas in realistic settings, funding is available for eligible project costs of:

• Up to 70% for a micro or small organisation
• Up to 60% for a medium-sized organisation
• Up to 50% for a large organisation