Poster Session B, Sunday, March 24, 8:00 – 10:00 am, Pacific Concourse
The role of alpha power as a stimulus-specific updating signal in sensory cortex post feedback in a reinforcement learning task
Khoi Vo1, Berry van den Berg1,2, Benjamin Geib1, Rene San Martin3, Marty Woldorff1; 1Duke University, 2University of Groningen, 3Centrode Neuroeconomia Universidad Diego Portales
Adaptive behavior is an outcome of successfully learning reward associations, whether for different stimuli or actions taken in an environment. Research on the neural mechanisms underlying the learning of these associations in humans has focused largely on the fronto-cortical/basal-ganglia reward circuits investigated with functional neuroimaging. However, few studies have focused on how sensory cortices may contribute to this learning phenomena, or on the within-trial time course of neural processes underlying such learning. Here we addressed this research gap by recording EEG while participants (n=30) performed a probabilistic foraging task for points by learning the stimulus-outcome associations of discrete categories of stimuli (faces and houses). Participants showed matching behavior over the course of learning, eventually choosing the richer-association category at its reward-probability rate. Attentional bias prior to choice, as measured by the N2pc ERP component toward the richer category, developed over the course of learning. Neural processing became stimulus-category specific following reward feedback, reflected by decreased alpha power over face-selective regions (reflecting more cortical activity) after winning-versus-losing following a face choice, but not following a house choice. Importantly, greater decreases in alpha power corresponded to larger prediction error (PE) signal recovered from a two-parameter reinforcement learning model. This modeling/brain-measure correlation supports the view that the category-specific decreased alpha power over face-selective regions represents a sensory-updating signal akin to a PE signal in reinforcement learning. These results thus support the view that sensory cortices are involved in learning stimulus-outcome associations, while also delineating the temporally specific processes by which they do so.
Topic Area: THINKING: Decision making