Time: 2.30 pm CET

Date: Tuesday, February 4.

Zoom: Meeting ID: 965 3682 5009
Passcode: 899758

Chair: Jozsef Fiser

Learning and representing probabilities in the human brain

Florent Meyniel

NeuroSpin (CEA-Saclay campus) and Institute for Neuromodulation (Sainte Anne Hospital), Paris, France

The brain has an internal probabilistic model of its environment that is useful for many aspects of cognition, such as decision making, planning, perception and social interactions. Learning, in particular statistical learning, is a key process by which the probabilities that make up this internal model are estimated. It is now well established that learning is an incremental process driven by surprising events (i.e. events that deviate from the expectations derived from the internal model). In recent years, it has become clear that the confidence (or, conversely, the uncertainty) associated with the estimation of this internal model is another key component of the learning process. I will briefly review behavioural, theoretical and neural (MRI, MEG) data suggesting that confidence regulates the learning process. I will argue that while the neural representations of these two key aspects of learning, surprise and confidence, are now reasonably well understood, the neural representations of what is being learned, the probabilities, remain quite elusive. I will report the results of a recent 7T fMRI study which suggests that probabilities are not linearly encoded in fMRI activity (as is the case for surprise and confidence, which covary with fMRI activity in many brain regions), but are instead encoded in fMRI activity in a highly non-linear manner.