Just got a nice visualization from PNNL researcher Sutanay Choundhury. It uses GraphLab spectral clustering to cluster different nodes in a network. And here is a paragraph about the meaning of the graphs:
The input graph was partitioned using the spectral clustering implementation in GraphLab. The size of the nodes in the rendered graph is
determined by a cost function. In this visualization, the cost function simply returns the degree of the node, although one may envision using centrality or other non-graph theoretic metric of importance. We used a threshold k (20 in this case), to display at most k nodes with highest costs in a cluster. The goal of the visualization was to provide a capability to see the distribution of "important" nodes across the clusters.
The above effort is a part from a larger project. Pacific Northwest National Lab, USA has a cyber security related project which uses GraphLab, among other tools. The project is called M&Ms4Graphs: multi-scale, multi-dimensional graph analytics for cyber security. Anyone who is interested in learning more about it, is welcome to attend our 3rd GraphLab conference, where Sutanay will give a demo.