A Density-Peak Approach to Clustering Graph-Structured Data

Clustering is the task of grouping elements in such a way that those belonging to the same group, or cluster, are more similar to each other than to those of other groups. The density-peak algorithm is a clustering procedure that is based on the idea that cluster centers have higher density than their neighbors, and that they are quite distant from points with higher densities. In this work, we exploit its flexible infrastructure and bring in some proper changes in order to use it for the important but complex task of clustering graph-structured data, i.e., a situation where the single elements to be grouped are networks. Potential implications include, but are not limited to, determining different groups of people based on their brain connectivity for ability assessment or disease prevention, as well as finding or confirming economic structures based on the different systems' networks. We validate our new method by comparing its performance on simulated data with that of other existing procedures, showing promising results. Finally, we discuss our main findings and outline possible future directions.