Fig. 4

(a) Population spike-count distributions \(P_{\mathrm{EIF}}(k)\) for the EIF model and \(P_{\mathrm{LNL}}(k)\) for the linear–nonlinear cascade approximation for 8, 32, 64, and 100 neurons for \(\mu= 0.1\) and \(\rho= 0.1\). While the distributions are very similar overall, the LNL model greatly overestimates the zero population spike-count probabilities and underestimates the tails. Inset: the same distributions on a log-linear scale. (b) The JS divergence between the EIF and LNL is an order of magnitude smaller than PME. (Also, the order of the mean firing rates is reversed when compared to the PME as the LNL cascade gives a better approximation at lower firing rates)