Controlled Analysis of neurocontrollers with informational lesioning Keinan Meilijson Ruppin 2003

This is an elaboration of functional contribution analysis (FCA), which is an approach to measuring the structure of dynamic networks. It ultimately assigns a vector to the network such that each entry is a measure of that node's contribution to the function of the whole network.

in FCA with informational lesioning, this vector is optimized to best predict the behavior of the network under all possible lesions. In practice it is not necessary to make all possible lesions (which would get intractable quickly), but only a small sample. The paper lays out all the procedures

For my own purposes, I can presumably measure the de/centralization of a system with something like the variance in the distribution of CVs.

Though I read this paper, it will deserve much closer scrutiny when I get into application/implementation.

This is the first/primary measure that Paul recommended to me when I asked about metrics for my robot work.