Figure 2.

Flow chart of the identification algorithm. The procedure starts (top left corner) with the stimulation of the neuronal network through the protocol ({Δν_ext}) described in the Section 2. From the resulting instantaneous firing rate ν(t), the time course of the adaptation dynamics c(t) is carried out for different guess values of τ_c. The estimated τ_c is satisfying an optimality criterium based on the anti-correlation between c and ν. The trajectories (c(t),ν(t)) are then employed in the generation of the fitting data for the vector field component of interest ({G(c,ν)} on the phase plane). These sparsely distributed values are subsequently interpolated using a thin-plate smoothing spline. The resulting function G(c,ν) defined over the whole phase plane, together with various reference values and a fitting model for Φ^(eff), allow to reconstruct both the effective gain function Φ^(eff)(c,ν) and the network relaxation timescale τ_ν(c,ν).