Jean-Claude Dreher is delighted to welcome Elsa Fouragnan from the Centre for Cognitive Neuroimaging at the University of Glasgow to come and give a talk:
Spatiotemporal characteristic of reward-based learning in humans
Adaptive decisions depend on accurate outcome representations associated with potential choices. These representations can be acquired with reinforcement learning mechanisms that use the prediction error (PE) – the difference between expected and actual outcomes – as a unique learning signal to update expectations. However, these signals are characterized by their valence and salience dimensions but their relative contributions as well as their underlying networks remain debated. Here, we coupled high temporal resolution, single-trial EEG with simultaneously acquired fMRI while participants performed a probabilistic reversal learning task and inferred the full spatiotemporal dynamics of the brain networks involved in reward-based learning. We identified two temporally and spatially distinct processing stages of outcome valence: an early process driven by an automatic alertness response to negative outcomes and a later, more deliberate, assessment of the value information required for updating value expectations. Importantly, these two valence systems interact to promote switching behavior via thalamostriatal interplay. Parallel to the late valence-evaluation, we found that the brain also represents quantitative information about PE salience, largely separate from the valence systems. Crucially, we found no evidence for a signed PE at a neural level but rather an overlap between the late valence and salience networks, showing a clear summation profile, making it unlikely that a unique learning signal exists at the cortical level in humans.
ISC - Council Room