Figure 6.
Computational model for how predictions influence neural signals corresponding to auditory listening and imagery. Auditory inputs elicit bottom-up S responses through the auditory cortex (ACX). A prediction model generates a top-down P signal that is more similar to S for more predictable sensory events. That prediction is compared with S, producing an error signal δsur = S−P. The EEG response is hypothesized to capture a combination of δsur and S, meaning that some level of EEG activation is expected even when S is fully predictable (Margulis, 2014). When a sound is imagined, S = 0, and therefore δsur = −P, as for our hypothesis in Figure 1.