Figure 8.

The classical implementation of predictive coding relies on the cortical hierarchy. A simplified hierarchical model based on the classical implementation of predictive coding (Bastos et al., 2012). Vertical dashed lines delimit hierarchically arranged cortical columns (left to right: bottom up). Prediction errors climb up the cortical hierarchy through the feedforward, bottom-up connections, whereas predictions are sent backwards (top-down) to suppress prediction error units of the levels below, via inhibitory connections. Superficial layers (L2/3) above the subcortical input layer (L4) carry prediction errors, whereas deep cortical layers (L5/6) carry predictions. Adapted, with permission, from Heilbron and Chait (2018) and Carbajal and Malmierca (2020).