Fig. 2

(a) Population spike-count distributions \(P_{\mathrm{EIF}}(k)\) for the EIF and \(P_{\mathrm{PME}}(k)\) for the pairwise maximum entropy (PME) model, for populations of \(N= 8, 32, 64\), and 100 neurons. Here \(\mu= 0.1\) and \(\rho= 0.1\) (input parameters \(\gamma=-60\mbox{ mV}\), \(\sigma=6.23\mbox{ mV}\), \(\lambda=0.30\)). The distributions \(P_{\mathrm{EIF}}(k)\) and \(P_{\mathrm{PME}}(k)\) are similar for smaller populations but differ larger populations. Inset: the same distributions on a log-linear scale. (b) The Jensen–Shannon (JS) divergence between the EIF and the pairwise maximum entropy (PME) model. We normalize by \(\log(N)\), the natural growth rate of the JS divergence. Left: JS divergence for a constant value of \(\mu= 0.1\) and increasing values of correlation ρ (input parameters \(\gamma =-60\mbox{ mv}\), \(\sigma=6.23\mbox{ mV}\), \(\lambda=0.17\), 0.30, and 0.59, respectively). Right: JS divergence for constant value of \(\rho= 0.1\) and increasing values of firing rate μ vs. population size. The firing rate was varied by increasing the DC component of the input current, γ (input parameters \(\sigma=6.23\mbox{ mV}\), \(\gamma=-60\mbox{ mV}\), −58.2 mV, and −56.8 mV, respectively, and \(\lambda=0.30\), 0.25, and 0.23, respectively). The JS divergence grows with increasing ρ and decreasing μ