The neural dynamics of feedforward and feedback interactions in predictive processing

Cortical areas are reciprocally coupled via feedforward (FF) and feedback (FB) connections that have distinct laminar profiles. Recurrent interactions between FF and FB streams may underlie context-dependent, flexible processing of sensory stimuli and the formation of predictions. Hierarchical predictive coding theories postulate that the communication of FF sensory input signals and FB predictions uses distinct frequency channels, namely gamma and alpha/beta-frequencies. Our review calls for an update of this dual-frequency theory: Empirical evidence suggests that both gamma and beta rhythms emerge during sensory and motor states characterized by high spatial and temporal predictability, in which case horizontal and top-down FB can explain away FF inputs. By contrast, broadband increases in gamma power for unpredicted stimuli likely result from increased firing rates. Recentevidence further indicates that distinct rhythms emerge in specific networks and are not consistently associated with either FF or FB influences, or with corresponding laminar compartments. Accordingly, we discuss potential functions and mechanisms of rhythms in the local circuit related to efficient coding and synaptic plasticity, rather than the inter-areal communication of FF error and FB prediction signals.