Fig. 8.

Conditional expectations of the coupling parameters: The upper panels show the conditional expectations of the coupling parameters plotted against their true values for the full model (left) and under the (optimum) reduced model (right). One can see the characteristic shrinkage of a subset of parameters to their prior mean; in this case the parameters associated with anti-edges (black dots). The veracity of this shrinkage depends upon the accuracy of model selection. The lower panel shows the same estimates; however, here, we have tried to highlight that the shrinkage to true values is also evident for the connections that were present (red dots). This panel plots the change in conditional estimate when moving from the full model to the selected (reduced) model against the coupling strength error under the full model. For the connections that were absent, this is simply a straight line, because in all cases, the correct model was chosen. Crucially, we see a similar effect for the connections which were present, with a mild negative correlation between the change in estimate and its mismatch under the full model. Inhibitory self-connections are not included in these results (and did not change much, because they were subject to relatively informative priors).