Figure 4. .

The figure shows a comparison of the model performances when varying the amount of data and the size of the network. The test MAE of the VAR is here depicted in orange, the MAE of the DCRNN in blue, and the error of the GWN in red. The overall error was computed as an average across brain regions, time steps and test samples. The error bars represent the standard deviation of the test error across subjects. (A) The test MAE using a dataset of 50 subjects is shown for the visual network, the network within the right hemisphere and the whole-brain network (Glasser et al., 2016). (B and C) The test performances in dependence of the network size using the 25 and 10 subject dataset, respectively.