Here's our attempt to break up Neurotree into 60 clusters and plot them in two dimensions. We treated the tree as a graph with nodes defined as people and edges defined as mentor relationships (each with equal weight). Clusters were genereated with a partitioning algorithm that uses spectral factorization (Hespanha, 2004) to minimize distortion as individual nodes are collapsed into aggregate clusters.
The cluster image map was then generated using Graphviz. Each cluster is represented by its member with the strongest average connection strength across the whole tree and by the two most common research areas within that cluster. Clusters are sorted from top to bottom roughly from oldest to youngest. Line weight indicates connection strength between clusters (connection count: dotted, 1-2; solid, 3-7; bold 8+). This page is updated every day or so.