CCN meeting | Florent Meyniel (INSERM-CEA Cognitive Neuroimaging unit, France)

21-11-2019 from 11:00 to 12:00
Henri Dunantlaan 2, room 4.3
Add to my calendar

Probability, learning and confidence in the brain

Humans possess a remarkable ability to learn an internal model of the world. This is all the more impressive that our world is changing and uncertain, making learning difficult. In this talk, I will present some properties of probability learning, which is essential to build an internal model. I will show that the learning of probabilities is fairly accurate, and accompanied by a sense of confidence, indicating the credence held by the learner in her probability estimates. This sense of confidence appears rational and shows several properties of a Bayesian inference. Confidence is not a gratuitous aspect of Bayesian inference, it has a specific function: the regulation of learning. I will show that surprising observations, which should prompt updating, and confidence, have distinct effects on brain dynamics measured with fMRI, MEG and pupillometry. Interestingly, confidence modulates the brain state, which in turn shapes surprise responses, resulting in confidence-weighted surprise responses suitable for Bayesian updating.