Aroma Dabas | Computational and neural basis of episodic-simulation induced learning

Since the 1970s, we have gained a decent understanding of how people learn from their experiences, picking up on contingencies between neutral conditioned stimuli (CS) and unconditioned stimuli (UCS). Specifically, such experience-dependent learning can be accounted for by reinforcement learning models. Recently, it has been shown that we not only learn from real experiences but also from episodes that we have merely imagined (Benoit, Paulus, Schacter, Nat. Commun., 2019). We here test the hypothesis that such simulation-induced learning is also governed by similar computational and neural processes. We thus have participants imagine experiences with personally familiar people (CS) (e.g. eating ice-cream with Sally on a sunny day). Critically, we manipulate whether these events are pleasant versus non-pleasant (UCS). Using this task, we track whether people develop a preference for CS that have been paired more frequently with pleasant events and whether this changes their overall attitude towards the CS (e.g., how much they like Sally). Using model-based fMRI, we will test the hypothesis that such learning is based on a striatal prediction error that updates value representations of the individual people in the medial prefrontal cortex.