Fig. 5

(a) Population spike-count distributions \(P_{\mathrm{EIF}}(k)\) for the EIF model and \(P_{\mathrm{DG}}(k)\) for the Dichotomous Gaussian (DG) approximation for 8, 32, 64, and 100 neurons for \(\mu= 0.1\) and \(\rho= 0.1\). The distributions are very similar, showing that the DG model very accurately captures EIF spiking statistics. Compare with Figs. 2, 4. Inset: the same distributions on a log-linear scale. (b) Left: JS divergence between the EIF and DG models for a constant value of \(\mu= 0.1\) and increasing values of correlation ρ; values appear noisy, but are several orders of magnitude lower than the JS divergence between EIF and PME or LNL models in Figs. 2(b), 4(b). Right: Similar, for a constant value of \(\rho= 0.1\) and increasing values of firing rate μ. The JS divergence grows with increasing μ