Study on the neural population response during visual expectation

Abstract

Prior expectation has shown to be engaged with changes of neural activity in the sensory areas which were essentially responsive to the physical input of sensory stimuli. Predictive coding theory delineates neural responses in visual cognition by reciprocal signaling between two computational units

representation (R) unit representing prior information of upcoming stimuli and error (E) unit encoding unpredicted portion between the prior information and the actual input. However, it is little known how cortical responses reflect visual expecta-tion in temporal dynamics. We used functional magnetic resonance imaging (fMRI) to measure the blood oxy-gen level dependent (BOLD) signal in humans while evoking blockwise category-specific expectation (low, mid, high probability of face stimuli). At the same time, the tasks were manipulated to be orthogonal with the expec-tations, so that participants did not attend on a specific category. Simulated BOLD timeseries were computa-tionally modeled by the weighted sum of representation and error signals, together constituting predictive code, where representation units develop to minimize the activity of error units. Compared to the traditional percep-tion model, predictive coding best explained the real BOLD signal in the category selective regions. ROIs in the left retrosplenial cortex and right frontal pole were correlated with representation signal, and left middle frontal gyrus was correlated with erro signal, indicating reciprocal signaling of two units across cortical hierarchy. These results suggest that neural population dynamics are determined by predictive coding during perceptual expecta-tion, and results support the idea that the predictive coding accounts for the neural mechanism in visual perception.