Figure 8
From: Towards an efficient validation of dynamical whole-brain models
Numbers of algorithm executions and corresponding probabilities to obtain a goodness-of-fit not smaller than 95% of that from the grid search in the 2Dim (A–D) and 3Dim (E–H) cases. The probabilities were evaluated by randomly selecting \(R \in \{ 1, \ldots ,R_{\max } \}\) goodness-of-fit values from the Rmax algorithm executions available for every subject (Rmax = 15 for PSO and CMAES, Rmax = 24 for NMA and BO). A success was noted when at least one of the selected values was above or equal to the threshold. For every choice of R (i.e. the number of performed runs indicated on the horizontal axes), this procedure was repeated 500 times. The results were then averaged across all subjects in order to obtain the mean success probabilities indicated on the vertical axes. For the optimization methods presented in the legends (same notations as before), the plots illustrate the mentioned success probabilities together with the respective standard error (error bars). Additionally, it is indicated after how many runs success probabilities of 50% and 80% could be surpassed. The figure was created with MATLAB R2018a (www.mathworks.com).
