New framework offers a better way to model complex networks
- By Kevin McCaney
- Jul 11, 2016
Networks have grown so large and complex that they’ve outstripped the ability of conventional modeling tools to accurately represent them.
Stanford University researchers funded by the Defense Advanced Research Projects Agency are taking an innovative approach to the problem, with a program that can identify patterns in complex networks that can be used in mathematical models to better represent those systems. The researchers say the framework moves recognition of connectivity from the lower order involving individual nodes to a higher order.
Such an approach could be used to better understand systems ranging from military logistics and social media to the interaction of molecules and proteins within cells and the operation of integrated circuits, DARPA said in a release.
“This approach mathematically represents complex networks more efficiently, revealing deeper functional relationships within networks and how each pattern contributes to the whole,” DARPA program manager Reza Ghanadan said in the release. “Additionally, it provides an analytic, systematic, and scalable way to generate hypotheses that are provably relevant to a given network based on key insights that the patterns reveal in that network. Taken together, this is an exciting demonstration of the promise that motif clustering shows for helping to unravel the complexity of diverse scientific and engineering systems, and for accelerating discovery by highlighting which avenues of research could potentially yield better results.”
As part of the agency’s Simplifying Complexity in Scientific Discovery, or SIMPLEX, program the Stanford researchers developed a framework for identifying “motifs”—sometimes obscure patterns—and grouping them together. In one example, they used the framework and its motif-clustering algorithms to automatically identify the eight largest hubs along air routes connecting the 50 most populous cities in the U.S. and Canada, something that’s not easily done with other methods. In a paper published in Science, the researchers say the framework can scale to networks with billions of edges.
Another team in the program, at Baylor University, is taking the same approach with protein networks in biological systems, which could lead produce better understanding of diseases, genome mapping and other medical benefits, DARPA said.
In addition to potentially helping the military better model its massive logistics systems—something it has aimed for in the past—the approach could be applies in other areas, such as monitoring social media activity for signs of extremism.
Kevin McCaney is a former editor of Defense Systems and GCN.