I think the authors are confused about what’s happening here. It’s not revealing a subnetwork, you’re removing spurious interactions in the fully connected ANN

I think the authors are confused about what’s happening here. It’s not revealing a subnetwork, you’re removing spurious interactions in the fully connected ANN thereby visually revealing the functional circuit topology that’s driving the behavior. It’s helpful to understand the effect of pruning on artificial gene regulatory network, another class of ANN but mathematically they follow the same rules: https://www.researchgate.net/publication/23151688_Survival_of_the_sparsest_Robust_gene_networks_are_parsimonious https://www.researchgate.net/publication/23151688_Survival_o...