The model has a rich dynamical repertoire (A, B) and can be correctly fitted to synthetic data (C-F).

<p>(A, B) The coefficient of variation (<i>C</i>_<i>V</i>) of the inter-spike interval distribution is computed for parameter values shown as black dots and then linearly interpolated. (A) The <i>C</i>_<i>V</i> of a simple version of the AGAPE (<i>k</i>(<i>t</i>) = <i>σ</i>²<i>e</i>^{− <i>t</i>/<i>τ</i>}, <i>α</i> = <i>η</i> = 0) as a function of the model parameters (membrane time-constant <i>τ</i>, baseline firing rate <i>r</i>₀ and coupling strength <i>βσ</i>). (B) <i>C</i>_<i>V</i> of the AGAPE model with an exponentially adaptive process with fixed membrane time-constant, firing rate and coupling (<i>βσ</i> = <i>r</i>₀<i>τ</i> = 1) as a function of the parameters describing adaptation, namely adaptation strength <i>η</i>₀ and time-constant <i>τ</i>_<i>r</i>. (C, D, E, F) Synthetic data is sampled from the AGAPE model with GP (D), spike-related (E), and adaptation (F) kernels as depicted in black, and <i>δ</i> = 4 ms, <i>r</i>₀ = 4.15 Hz, <i>β</i> = 0.374 mV^{− 1}. Then the AGAPE is fitted to the synthetic data by maximum likelihood (ML). (C) The maximum log likelihood per bin as a function of the parameter <i>δ</i> has its maximum at the ground truth value <i>δ</i> = 4 ms. (D, E, F) The ML estimates (red) of the GP, spike-related and adaptation kernels lie within two standard deviations (red shaded regions, estimated by means of the observed Fisher information) from the ground truth.