A neural representation of prior information during perceptual inference

Perceptual inference is biased by foreknowledge about what is probable or possible. How prior expectations are neurally represented during visual perception, however, remains unknown. We used functional magnetic resonance imaging to measure brain activity in humans judging simple visual stimuli. Perceptual decisions were either biased in favor of a single alternative (A/similar to A decisions) or taken without bias toward either choice (A/B decisions). Extrastriate and anterior temporal lobe regions were more active during A/similar to A than A/B decisions, suggesting multiple representations of prior expectations within the visual hierarchy. Forward connectivity was increased when expected and observed perception diverged ("prediction error" signals), whereas prior expectations fed backward from higher to lower regions. Finally, the coincidence between expected and observed perception activated orbital prefrontal regions, perhaps reflecting the reinforcement of prior expectations. These data support computational and quantitative models proposing that a visual percept emerges from converging bottom-up and top-down signals.