Networks have two great virtues: They can describe interactions between lots of objects, and those objects can be different. No classical theory meets both of those challenges.
But merely describing something as a network doesn’t gain you much. Plot almost any complex network and you’ll end up with something that looks like a plate of spaghetti. Realizing the potential of network theory requires building new tools to sort that spaghetti.
Three SFI researchers have won a three-year, $417,000 grant from the James S. McDonnell Foundation to attempt to do just that.
The best way to understand networks, the collaborators have found, is to get computers to figure out the networks for themselves. SFI Professor Cris Moore (University of New Mexico), Postdoctoral Fellow Aaron Clauset, and External Professor Mark Newman (University of Michigan) have been applying machine learning tools to networks, with powerful results.
For example, Cris’s student Tiffany Pierce studied the network formed by all the words in the novel David Copperfield, with two words being connected if they appear next to each other.
“We told the computer, go and find an explanation of this network in terms of three types of vertices,” Cris says. “We didn’t tell it anything else.”
The machine learning algorithms managed to sort them into nouns, adjectives, and words like “light” that can play both roles.
Now the team wants to extend its work to more complicated situations, for example when the edges connecting nodes are of varying types, or where their locations play a crucial role in how they connect.