Abstract
Expectations about the sensory environment have a strong impact on perceptual decisions. However, the computational and neural mechanisms by which perceptual expectations are updated, maintained and used for visual perception remain poorly understood. In this model-based fMRI experiment on 15 healthy participants, we used an associative learning task to elicit changing expectations about the appearance of ambiguous stimuli. Conventional analyses revealed that perceptual decisions under ambiguous viewing conditions were biased by perceptual history and learned associations. In a computational modeling approach, we showed that participants' behaviour was best explained by a hierarchical Bayesian model incorporating continuously updated priors from previous perceptual decisions and associative learning. Model trajectories for predictions and predictions errors correlated with BOLD-activity in orbitofrontal gyrus and inferior frontal gyrus, respectively. Effective connectivity analyses using Dynamic Causal Modeling indicated a relevant impact of such regions on activity in sensory cortices via feed-back connections. Our results suggest that visual perception is informed by different sources of prior expectations, which are updated by Bayesian learning and used to infer on the causes of sensory stimulation in a continuously changing environment.
Meeting abstract presented at VSS 2017