Our work focuses on how neurons in our brains build models of the relationships between objects and events in the world to allow flexible behaviour.
This raises interesting computational problems, such as how best to organise knowledge of the world. It requires rich tasks that are most easily developed in humans, but it requires solutions expressed at the level of neuronal representation that is most easily studied in animal models.
Through cross-species investigations, we relate large-scale measurements that can be made in humans to cellular measurements and direct interventions that can be made in non-human species.
Theoretical understanding of neuronal representations in hippocampus and frontal cortex
One of the few places where we do have a detailed understanding of how neuronal activity produces flexible behaviour is open-field spatial navigation (leading to John O’Keefe from UCL winning the 2014 Nobel prize). We have shown that these same cellular computations also represent relationships outside of spatial reasoning. We are developing new theoretical frameworks that generalize these representations to arbitrary (non-spatial) relational reasoning. They provide a formal bridge between two domains of neuroscience (spatial cognition and reinforcement learning) and therefore, new opportunities to understand neural computations in both domains.
New animal experiments to test these theories
In collaboration with Neil Burgess and other researchers at UCL (Barry, Caccucci) and Oxford (Walton, Akam), we have a major new endeavour to build rodent assays to test these theories. They will provide a substantial neuronal dataset from hippocampus and frontal cortex in both spatial and non-spatial tasks. Critically these tasks are extremely rich by comparison to usual rodent assays, yet under precise mathematical control. Therefore, they permit strong tests of representational theories. Simultaneously, with Steve Kennerley, we are developing new macaque behavioural assays where animals can build knowledge structures that contain hundreds of elements.
New tools to index these representations in humans
We are developing tools using fMRI and MEG to index these representations and computations in humans. For example, with Ray Dolan, we have developed new ways of measuring replay with MEG. We have shown that this replay shares many properties with rodent replay, but also that it can perform functions much more sophisticated than have been reported in rodents. It recombines knowledge to represent new potential experiences. It re-orders events to fit with prior knowledge. These rapid subconscious events are likely a key mechanism for flexible cognition.
- The human brain reactivates context-specific past information at event boundaries of naturalistic experiences Nature Neuroscience
- How do the blind ‘see’? The role of spontaneous brain activity in self-generated perception Brain: a journal of neurology
- View all publications by the Decision Making team