Figure 9

Correlational structure of dynamics and connectivity features in optimised networks. (a) An example of the evolutionary process. Synaptic weights reciprocity (r) and transition graph modularity (Q) are shown, computed for the elite individual in each generation of a 50 generations evolutionary process. Modularity was the fitness function in this case. Pearson correlation coefficient (CC) between both measures is shown, together with its p-value. ((b)) Transition graph \(G_{cons}\) obtained with gFTP from the elite individual in the last generation of the evolutionary process shown in (a). Nodes plotted with different shapes (circle, square, triangle) indicate the three modules that maximized modularity. \(Q=0.46\), \(r=0.19\) for this graph and its associated network, respectively. (c) Correlation matrix between 8 measures that quantify distinct aspects of network dynamics and connectivity, obtained from 20 independent repetitions of the evolutionary processes. Measures computed were: neuron number (\(N_{neu}\)), node number (\(N_{v}\)), information between network state and stimulus (I), modularity (Q) and clustering coefficient (c) of the transition graph, reciprocity (r), absolute reciprocity (\(r_{abs}\)) and outward strength variability (\(\sigma _{out}\)) of the synaptic weight matrix (see “Methods” for details on each measure). Measures were always computed on the elite individual, obtaining one matrix for each repetition and for each fitness function. Each panel shows the correlation matrix, averaged across repetitions, for the fitness function indicated in the panel title. The matrix shown in the last panel is the average over all the other matrices. Colour bar in first panel shows colour scale for all panels (pure red for CC = 1, and pure blue for CC = − 1).