Schematic representation of the HMAX-like Bayesian network.

<p>The 5-layer Bayesian tries to replicate the structure and functionality of a simplified version of the HMAX model <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0048216#pone.0048216-Serre1" target="_blank">[37]</a>. The probability distribution of each Bayesian node (grid square) represents the sum-normalized response of HMAX units at that location and layer, where the states of the node represent the different features (e.g. four Gabor filters). The conditional probability tables linking the nodes of different layers serve to approximate the HMAX selectivity and invariance operation (see text for details). The number of nodes per layer and the number of states per node is indicated beside each layer. The downward arrows between the layers (square grids) indicate the causal dependencies in the Bayesian network, e.g. C1 nodes are the parents of S1 nodes and the children of S2 nodes. The specific connectivity patterns between nodes are shown in <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0048216#pone-0048216-t001" target="_blank">Table 1</a>.