Fig. 3.

Belief propagation and intrinsic connectivity. This schematic features the correspondence between known canonical microcircuitry and the belief updates in Fig. 2. Left Panel: a canonical microcircuit based on (Haeusler and Maass, 2007), where inhibitory cells have been omitted from the deep layers – because they have little interlaminar connectivity. The numbers denote connection strengths (mean amplitude of PSPs measured at soma in mV) and connection probabilities (in parentheses) according to (Thomson and Bannister, 2003). Right panel: the equivalent microcircuitry based upon the message passing scheme of the previous figure. Here, we have placed the outcome prediction errors in superficial layers to accommodate the strong descending (inhibitory) connections from superficial to deep layers. This presupposes that descending (interlaminar) projections disinhibit Layer 5 pyramidal cells that project to the medium spiny cells of the striatum (Arikuni and Kubota, 1986). The computational assignments in this figure should be compared with the equivalent scheme for predictive coding in (Bastos et al., 2012). The key difference is that superficial excitatory (e.g., pyramidal) cells encode expectations of hidden states, as opposed to state prediction errors. This is because the prediction error is encoded by their postsynaptic currents, as opposed to their depolarisation or firing rates (see main text). The white circles correspond to the Bayesian model average of state expectations, which are the red balls in the previous figures (and the inset). Black arrows denote excitatory intrinsic connections, while red arrows are inhibitory. Blue arrows denote bottom-up of ascending extrinsic connections, while green arrows are top-down or descending (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article).