Figure 6
Correlation and reciprocity differentiate networks with sequence memory from random transition networks. (a–c) Networks constructed to solve the s-task with \(\tau = 7\) and \(f_{r} = 1\) (T + F networks), their isofunction network (T network), and networks with the same number of neurons and network states that follow a random transition graph (F networks). (a) Correlation increases as FR approaches 0.5 spikes/time step. F networks show a positive offset in comparison with T + F and T networks. (b) The dependency between reciprocity and FR is similar to the dependency between CC and FR. Higher reciprocity values are found in F networks. (c) Reciprocity grows linearly with correlation, as expected from panels (a) and (b). (d–f) idem a–c, but with \(f_{r} = 4\). (d) The CC/FR relationship is similar to the one observed with a lower neuron number (panel (a)). (e) The reciprocity/FR relationship inverted as the neuron number was increased. Reciprocity is minimized as FR approaches 0.5 spikes/time steps. F networks show pronounced negative reciprocity. (f) Reciprocity decreases linearly with correlation, as expected from panels (d) and (e). Mean ± SD are shown; n = 10 networks were constructed for each target FR and network type. All random graphs were constructed with \(f_{bc} = 0.25\).
