Network architecture and parameters.

<p><b>A</b> Two neural populations with rates <i>r</i>₁, <i>r</i>₂ inhibit each other with synaptic efficacies −<i>J</i>. <b>B</b> Two coupled balanced subnetworks, each consisting of an excitatory and inhibitory population. Excitatory (inhibitory) connections are represented in blue (red). Mutual inhibition is generated by all-to-all connections of strength from each inhibitory population to the excitatory population of the other subnetwork. As in [<a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1005505#pcbi.1005505.ref007" target="_blank">7</a>], connections within each subnetwork are random with a connection probability <i>K</i>/<i>N</i>, 1 ≪ <i>K</i> ≪ <i>N</i>. Connection strengths are: and according to the identity of the participating neurons. Without loss of generality, we chose <i>J</i>_<i>EE</i> = <i>J</i>_<i>IE</i> = 1 and define <i>J</i>_<i>EI</i> ≡ −<i>J</i>_<i>E</i>, <i>J</i>_<i>II</i> ≡ −<i>J</i>_<i>I</i>. An excitatory input is fed into both excitatory populations.