FIGURE 2.

Biologically motivated realization of deep Hebbian predictive coding. Each rectangle denotes a population of neurons that represents a specific signal, computed in predictive coding. The populations that compute errors are denoted by red blocks and the populations that represent inferred causes are denoted by blue blocks. Arrows represent excitatory connections and filled circles denote inhibitory connections (note that inhibitory interneurons were not explicitly modeled here). The interareal connections between representation neurons and error neurons are plastic (Equation 8) whereas the intra-areal connections are not. Connections conveying information required for the inference and learning steps of predictive coding are shown as black lines and other connections are shown in gray. Gating mechanism has not been shown in this figure for simplification. See main text for explanation of symbols.