Regularization for graph (neural) network structure learning

Question

I am looking for a regularization function that takes the number of connections into account so that dense networks are more punished than sparse networks. I found that for