Limitations of This Work

There are clear limitations of this work mostly caused by the lack of computational power available. We were not able to test a network representing the null hypothesis (a random DAG \cite{karrer2009random}) for this experiment (our runtime kept crashing when trying to run analysis on it). We would also like to have run the full suite of metrics on a full sized neural network (e.g. the MNIST network we describe in Section \ref{449423})  instead of a toy one and on large scale natural analogs like the mouse visual cortex connectome. In order to do so we would most likely need to implement a more powerful analysis suite using something like a compute cluster or GPU computing. We also only examined 4 classes of networks, to make a larger claim about the nature of collaborative networks we would need to analyze many more.