Figure 6
Only \(\approx 6\%\) of removed edges lead to significant qualitative changes in the network’s behavior (from (b) to (c)). The input applied to the original all-to-all network does not qualitatively change the collective behavior (from (b) to (a)). The trained network without any external input resides in a vicinity of the critical state (c) and reaches the criticality under the external input (d). This is additionally demonstrated in Fig. 7. Nodes’ labels on the graph (e) indicate the difference between the number of the node’s outgoing edges for the initial and the trained topology. These removed edges are highlighted with red color.
