Skip to main content
Go to the home page of the European Commission (opens in new window)
English en
CORDIS - EU research results
CORDIS

Neural algorithms and representations of flexible behaviour

Project description

Unravelling the brain’s code for flexible thinking

Humans can figure out new situations by drawing on patterns they have seen before. We can plan, adapt, and reason flexibly, whether it is knowing to wrap a gift before opening it or understanding that only registered voters can cast a ballot. However, scientists know little about how the brain supports this kind of thinking. Research so far has focused on spatial memory and the hippocampus. With this in mind, the ERC-funded NARFB project seeks to uncover the neural algorithms at work. Combining computational models, theory, and neuroscience, it will map how different brain regions work together to enable generalisation. The findings will offer fresh insight into both human cognition and artificial intelligence.

Objective

Humans and other animals learn from experience and generalise it to new situations. This generalisation permits rich inferences from sparse data and building novel plans in the face of new obstacles. Yet our understanding of the neural mechanisms underpinning flexible behaviours is limited to simple situations such as spatial learning and only in the hippocampal formation. Flexibly behaviour and generalisation is much more than just space and the hippocampal formation – it is for any structured knowledge. Flexible generalisation lets you know that presents should be wrapped before being opened, in the same way as knowing that you can only vote if you are already registered. Such structured knowledge, and their corresponding tasks, are often hierarchical, compositional, and require inferences., yet we have little understanding of the neural mechanisms of hierarchical, composition, or inference. This grant aims to build an understanding of the underlying neural algorithms and representations permitting these cognitive functions. By combining modelling and theory, we will investigate the optimal neural representations and algorithms for solving these tasks and relate them to prefrontal cortex and the hippocampal formation. We will use recurrent neural networks trained to learn and generalise on structured tasks. Furthermore, we will understand these systems theoretically - providing a mathematical formalism for optimal neural representations and algorithms that can explain when and why neural systems learn different representations and algorithms. We will collaborate with experimentalists to test model and theory predictions. Lastly, we will use the understanding gained from neuroscience tasks to build bridges between natural and artificial intelligence. Building such an understanding will offer deep insights into the neural mechanisms of cognition, how and when it dysfunctions, and allow us to interpret modern AI systems that have an increasing impact on humanity.

Fields of science (EuroSciVoc)

Programme(s)

Topic(s)

Funding Scheme

HORIZON-ERC - HORIZON ERC Grants

See all projects funded under this funding scheme

Call for proposal

(opens in new window) ERC-2025-STG

See all projects funded under this call

Host institution

THE CHANCELLOR, MASTERS AND SCHOLARS OF THE UNIVERSITY OF OXFORD
Net EU contribution
€ 1 486 863,00
Address
WELLINGTON SQUARE UNIVERSITY OFFICES
OX1 2JD Oxford
United Kingdom
Region
South East (England) Berkshire, Buckinghamshire and Oxfordshire Oxfordshire
Activity type
Higher or Secondary Education Establishments
Links
Total cost
€ 1 486 863,00

Beneficiaries (1)