There are two compatible perspectives on learning. It can be viewed as a dynamic phenomenon, where the reorganisation of behaviour is driven by nonlinear adaptations of neural activity, or as a stochastic, nonstationary process with interacting behavioural and neural components. In this talk, I will present statistical approaches for identifying low-dimensional nonlinear dynamics from high-dimensional neural activity during learning. I will show that reward-mediated learning in rats, rather than being a gradual improvement in the animal’s performance, manifests as rapid transitions between discrete behavioural states. Changes in behaviour are driven by coordinated neural activity in the prefrontal cortex that are signatures of a small number of stable attractor states. Most important, this low-dimensional neural representation can explain behavioural idiosyncrasies that cannot be accounted for using classical approaches for relating neural dynamics to behaviour.
A recording of this talk is available here: https://Universityofexeter.zoom.us/rec/share/VK4G4X75tQpwb4si8LTbnrx-ysxKPsIVvvSBPGx6GV6l5we9qrihjrRdKnS4SYd-.__V7VXb070Xt_qn8
Please email Kyle Wedgwood at email@example.com for access to the recording.