In order to act adaptively, we are remarkably capable in combining various sources of information to guide our behaviour. We learn from valenced experience, incorporate information using inference, and utilise errors to readjust information processing via increases in cognitive control.
In my talk, I will show how valenced and unvalenced outcomes are differentially processed in the human brain using EEG. These processes can additionally predict future choices on a single-trial basis. I will also relate predictions from sequential sampling models of choice formation with beta power lateralization (BPL). Within this framework, BPL was a reliable neural marker of choice formation. BPL further allowed to test whether decision boundaries are time-invariant or dynamically adapted on a trial-wise basis. BPL measurements could also be used to test whether the brain uses errors in an adaptive fashion to quickly increase performance, or if errors trigger an orienting response with maladaptive immediate consequences.
While the first part is focused on moment-to-moment decision making for valenced and non-valenced choices, the second part will address the interaction between valence processing and inference using a novel two-urn task and fMRI. This task allowed to behaviourally and neurally differentiate reward-based (model-free) learning, and inference (model-based learning) on a trial-by-trial basis.
Overall, my studies aim at a better understanding of how error-processing, simple valence learning and higher-level inference develop over time and interact to drive successful adaptive behaviours.