So much of the information in our world is networked, with nodes of data linked together through complex relationships. Just think of Facebook, where each user is a node connected to thousands of other users. But while the internet is built on complicated networks–and they’re incredibly relevant to how data is organized in the 21st century–they’re very, very difficult to visualize or understand.
A project called WonderNet maps complex networks in three dimensions using a new technique called the Elastic Link model. Created by a team of researchers from the Barabási Lab at Northeastern University and IBM’s Visual AI Lab, WonderNet transforms databases into data sculptures, all of which take different forms based on the underlying structure of the data. Unlike previous modeling techniques, which have trouble transforming data sets into physical objects in such a way that none of the links between data points overlap, the Elastic Link model finds an organic route for every single link.
“You can put now thousands of nodes and links, a random network, inside a box,” says Mauro Martino, the director of IBM’s Visual AI Lab and a collaborator on the project. “Activate the [Elastic Link] system, and it finds the best possible configuration without the overlap of nodes and links in a 3D space.”
For Martino, whose job at IBM is focused on finding ways to make AI understandable, the technology could have huge consequences for visualization: If you can plot the layout of a network in 3D space, you have a better shot of understanding exactly how it works. And because neural networks are based on vast amounts of data, Martino thinks that this kind of technology could make it easier for researchers to understand how AI makes decisions. For now though, he’s starting with transforming simpler networks into gorgeous visualizations. “It becomes kind of a sculpture,” Martino says. “It’s finally really readable.”
So far, the Elastic Link technique has been applied to a series of eight databases, ranging from one that details the relationships between artists and galleries all over the world to one that maps the citations between a host of academic disciplines. Another looks at how fake news spreads on Twitter. Each visualization is strikingly different, and for good reason. These maps can reveal much about the macro-level organization of how the networks function.
For instance, the visualized network of academic disciplines has a ring-like structure, showing just how interrelated all the disciplines are. The fake news map looks almost like a mushroom: Martino explains that the host of nodes on the head of the mushroom represent all the bots created during the Pizzagate scandal, all of which target a single giant node in the middle of the map–an influential person, who then slowly begins to believe the bots and spread the fake news out into the real-news ecosystem.
Along with the 3D maps, the team also created 360 videos to immerse viewers in the sculptures completely, and 3D-printed sculptures that show each network in real life. Martino and the rest of the researchers are also working on a system to predict how networks will behave, and plan to release the Elastic Link technology to the public for designers to work with later this year. He has high hopes for WonderNet’s website: “This will be a place in the future to go to find the best possible visualization of networks,” he says. For now, you can play around with each of the eight visualizations online here.