Design of Collaborative Networks

The design of many collaborative systems up has a lot of literature behind it, from observing successful organizations, the way natural systems interact and other existing systems, academia has come up with many theories on effective structures for collaboration. Often these have be summed up in common rules of thumb, like the importance of centralized decision making\cite{zhang2011designing}, the size of teams.\cite{smith2017maintaining} or adages like C-Level executives having an open door policy to improve productivity and moral\cite{detert2007leadership}.  Once a system has been created, it is also common for these networks to slowly refine themselves based on performance metrics  (e.g. output, revenue, survival). For naturally occurring collaborative networks, these systems have been refined over generations through natural selection. While this means that over time the efficiency of collaborative systems is improving, there may yet be techniques and structures unexplored due to the nature of iterative refinement. 

Topological Metrics to Aid in the Design of Collaborative Networks

We in this work,  examine the potential for creating topological metrics to aid in the design of collaborative networks so that instead of relying purely experience, intuition and refinement based on performance, designers of these networks have solid design parameters to tune for the results they are looking for. The features of networks are numerous and present many different tunable parameters. 
We have chosen to examine the role of bottlenecking and betweenness centrality, to which we compare another centrality metric - PageRank. In addition we also examine in/out degree distribution and all pairs max flow on a variety of networks each exhibiting a different level of effectiveness at collaborating. Using these metrics we hope to see qualities that emerge differentiating effective collaborative networks from the rest. The ability to show that these metrics signify effective collaboration could lead to a new set of tools to help designers in the creation of new networks.  

Related Work

The metrics we have chosen to examine stem from evidence shown in the literature, notable cases include the evaluation of protein networks and of various artificial neural nets lead us to believe that bottlenecking signifies effective collaboration

The Role of Bottlenecks in Protein Regulation Networks

Yu et al. \citealt{yu2007importance} show through an analysis of protein regulation networks in yeast that bottlenecks in the network are correlated with essential proteins. This was known about protein networks before Yu et al.'s work however due to analysis approaches it was not clear if the correlation was due the degree of nodes in the protein network or the betweenness of nodes. Yu et. al shows that it is indeed a factor correlated with betweenness in particular. 

Development of Specialized Nodes and Sub-Graphs and Their Importance in Artificial Neural Nets